Machine learning approach for differentiating iron deficiency anemia and thalassemia using random forest and gradient boosting algorithms

Machine learning approach for differentiating iron deficiency anemia and thalassemia using random forest and gradient boosting algorithms | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Machine learning approach for differentiating iron deficiency anemia and thalassemia using random forest and gradient boosting algorithms Wanicha Tepakhan, Wisarut Srisintorn, Tipparat Penglong, Pirun Saelue This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5623304/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 15 May, 2025 Read the published version in Scientific Reports → Version 1 posted 8 You are reading this latest preprint version Abstract Formulas based on red blood cell indices have been used to differentiate between iron deficiency anemia (IDA) and thalassemia (Thal). However, such formulas exhibit varying efficiencies. In this study, we aimed to develop a tool for discriminating between IDA and Thal by using the random forest (RF) and gradient boosting (GB) algorithms. Complete blood count data from 1,143 patients with anemia and low mean corpuscular volume were collected (382 patients with IDA, 635 with Thal, and 126 with IDA and Thal). The data were randomly divided into training and testing datasets by using a ratio of 80:20. The RF and GB models had good diagnostic performances for predicting IDA and Thal in the training and testing datasets. In the testing dataset for predicting binary outcomes, GB and RF both had an accuracy of 90.7% and an area under the receiver operating characteristic curve (AUC-ROC) of 0.953. A lower diagnostic performance was observed when patients with IDA and Thal were included. GB and RF showed accuracies of 80.4% and 82.2%, respectively, and AUC-ROC values of 0.910 and 0.899, respectively. A machine learning approach was developed using GB algorithm. This tool may be useful in regions where Thal and IDA are endemic. Health sciences/Diseases/Haematological diseases/Anaemia Health sciences/Medical research/Experimental models of disease iron deficiency anemia thalassemia machine learning random forest gradient boosting Figures Figure 1 Figure 2 Introduction Anemia is a common condition encountered in clinical practice. It is defined as a low number of red blood cells (RBCs) or a low hemoglobin (Hb) concentration. Anemia is classified into three categories on the basis of the Hb concentration and RBC size: hypochromic microcytic anemia, normochromic normocytic anemia, and macrocytic anemia [ 1 ]. Iron deficiency anemia (IDA) and thalassemia (Thal) are the most common causes of hypochromic microcytic anemia. A 2021 global survey reported that the prevalence of anemia was 24.3%, and approximately 66.2% of the total anemia cases are caused by IDA [ 2 ]. IDA is characterized by a depleted iron storage that leads to low RBC production. Moreover, it can be caused by a low iron intake, acute or chronic blood loss, or abnormalities in iron absorption. The prevalence of IDA is approximately 1.5–12% in the Thai population [ 3 , 4 ]. Thal is one of the most common causes of anemia [ 2 ]. It is an inherited disorder caused by a mutation in the globin gene that results in reduced or absent globin chain production. Patients with Thal traits (TTs) usually exhibit no anemic symptoms. By contrast, patients with Thal disease show widely different clinical phenotypes (from mild to severe anemia) depending on the mutation type. In Thailand, the prevalence rates of TTs, including α-TT, β-TT, and heterozygous Hb E, are approximately 20–30%, 3–9%, and 10–50%, respectively [ 5 ]. Laboratory investigations for diagnosing these conditions include serum iron tests, ferritin level assessment, Hb analysis, and deoxyribonucleic acid (DNA) analysis for Thal [ 6 , 7 ]. However, Hb and DNA analyses are unavailable in some hospitals owing to the need for specialized equipment, advanced technical expertise, and the time-consuming nature of the tests. In addition, these investigations can be costly, as patients often incur substantial expenses when physicians request comprehensive confirmatory tests for the diagnosis. Therefore, several mathematical formulas based on RBC indices, including Sirdah, Green and King, Mentzer, England and Fraser, Ehsani, Srivastava, Shine & Lal, and the 11T score, have been developed to help clinicians select appropriate confirmatory tests for differentiating between IDA and TTs, aiming to reduce investigation costs and time [ 4 , 8 ]. However, the efficiency of these equations varies. Moreover, the cut-off values of each formula are affected by sex, age, and ethnicity, resulting in unsatisfactory sensitivity and specificity results among different populations [ 9 , 10 ]. Applying these formulas for the differential diagnosis between IDA and Thal, including TT and Thal intermedia (TI), offers limited diagnostic value. In recent years, several machine learning algorithms such as C4.5 decision tree, k-nearest neighbor, artificial neural network, support vector machine, Naive Bayes, random forest (RF), vote algorithm, and extreme learning machine were evaluated for their classification performance in predicting whether a patient has IDA or TT [ 11 – 13 ]. RF showed a high performance with accuracies of 94.17% and 96.0% for IDA and TT, respectively [ 11 , 12 ]. In addition, the gradient boosting (GB) algorithm has been an effective model for predicting and diagnosing several diseases [ 14 ]. However, this algorithm has limited information in discriminating between patients with IDA and Thal. Thus, owing to the advancements of machine learning algorithms and the limitations of previous formulas for the differential diagnosis of IDA and Thal, this study aimed to generate a diagnostic model by using RF and GB algorithms to predict the probability of IDA and Thal. The results of the study should aid clinicians in determining appropriate laboratory investigations for patients with hypochromic microcytic anemia. Methods Participants This cross-sectional study was conducted at Songklanagarind Hospital, the largest tertiary hospital in southern Thailand, between January 2015 and December 2019. We assessed the first-visit data of 7,488 patients, who had the following characteristics: 1) age >15 years, 2) Hb concentration <13 g/dL in men and menopausal women or <12 g/dL in reproductive women, 3) mean corpuscular volume (MCV) <80 fL, 4) available iron profiles and ferritin level data, and 5) Hb and DNA analyses for Thal. Patients with anemia of inflammation, transfusion-dependent Thal, pregnancy, or incomplete laboratory data were excluded. To exclude anemia due to inflammation and pregnancy, a hematologist reviewed the medical records to confirm the diagnoses of IDA and Thal and to exclude patients with inflammation and infection. Patients with serum ferritin levels <30 ng/mL and transferrin saturation <16% were diagnosed with IDA [15]. All patients were diagnosed with Thal (TT and TI) by using the following diagnostic criteria: patients with Hb type A2A and Hb A2 levels ≥3.5% were diagnosed with β-TT. Those with Hb type A2A, Hb A2 levels 10%–35% were considered to have the Hb E trait. Patients diagnosed with TI exhibited Hb patterns such as A2FA, EFA, EE, A2AH, A2ABart'sH, CSA2AH, CSA2ABart'sH, EABart's, EFABart's, CSEABart's, and CSEFABart's; furthermore, these patients had no history of transfusion, and their conditions were confirmed through DNA analysis. The definitions and full names of the abbreviations are shown in the appendix. Patients who met the criteria for IDA and Thal were diagnosed as having IDA with Thal. Ethical approval Ethical approval for this study was obtained from the Human Research Ethics Committee (HREC) of the Faculty of Medicine, Prince of Songkla University (REC 62-232-5-2). The HREC waived the requirement for informed consent because this study used deidentified data. Laboratory techniques The hematological features were measured using an automated blood cell counter (XN3000; Sysmex Corp., Kobe, Japan). Hb analysis was performed using capillary electrophoresis (CapillaryS2; Sebia, Lisses, France). Serum iron levels, total iron binding capacity, and ferritin levels were measured using an automated analyzer (Cobas e411; Roche, Rotkreuz, Switzerland). DNA analysis for Thal was performed using polymerase chain reaction and reverse dot blot hybridization, as previously described [7, 16]. Statistical analysis The baseline characteristics and hematological features of patients with Thal and IDA were compared using Pearson’s chi-squared test for categorical data and the Kruskal–Wallis rank sum test for continuous data. P < 0.05 was considered statistically significant. The complete blood count (CBC) data of patients diagnosed with Thal (TT and TI), IDA, and IDA with Thal (TT and TI) were divided into training and testing sets by using a ratio of 80:20. Nine features were used in the machine learning methods, including Hb levels, hematocrit (Hct), MCV, mean corpuscular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), red cell distribution width (RDW), RBC count, age, and sex. Two types of models were built: binary outcome models (Thal and IDA) and multiclass outcome models (Thal, IDA, and IDA with Thal). Two ensemble machine learning classification methods were used to diagnose Thal and/or IDA.RF builds multiple decision trees on random subsets of the training dataset. This reduces the correlation between trees and avoids overfitting by selecting random subsets of features at each split. The final prediction is the majority vote from all the trees. GB builds trees sequentially by using information learned from previous trees. In theory, this improves accuracy compared to single-decision trees [17]. The model features were optimized using 10-fold cross-validation to maximize the area under the receiver operating characteristic curve (AUC-ROC). The Latin hypercube method was used to sample 1,000 sets of parameter values for each model. The best features were those that maximized the AUC-ROC of the validation data. The performance on the test set was evaluated using a range of metrics, including accuracy, kappa coefficient, sensitivity, specificity and AUC-ROC. To address the class imbalance, the synthetic minority over-sampling technique [18] was employed to generate synthetic samples for the minority class, thus effectively balancing the dataset prior to model training. A comparison between the diagnostic performances of GB and RF with formulas based on RBC indices such as Hct/Hb, MCV/Hb, Keikhaei, Jayabose, Sirdah, Green and King, Mentzer, England and Fraser, Srivastava, Shine & Lal, Matos, Ricera, Kerman I, Kerman II, Ehsani [8, 19-22] was made by using the method proposed by Delong et al. [23]. The definitions of all features and formulas are shown in the appendix. Analysis was performed using R version 4.4.2 [24]. The model specifications and analytical processes were performed using tidymodels version 1.2 [25]. The underlying analytic packages for RF and GB were ranger version 0 . 17 . 0 [26] and xgboost version 1 . 7 . 8 . 1 [27], respectively. The over-sampling of the minority class was performed using themis version 1.0.3 [28]. AUC-ROC was calculated using pROC version 1.18.5 [29]. Results A total of 14,407 CBC records of patients with anemia and low MCV from 2015 to 2019 were assessed. First-visit data from 7,488 patients were selected. In total, 6,345 patients with anemia of inflammation (n = 1,535), transfusion-dependent Thal (n = 65), or incomplete laboratory records (n = 4,745) were excluded. Therefore, 1,143 patients were included in the study. The data were randomly divided into two sets, namely, the training and testing datasets, by using a ratio of 80:20. Figure 1 shows the distribution of the cases in the training and testing datasets. Table 1 summarizes the baseline characteristics of the study groups. All RBC indices differed significantly among the three groups. Supplementary Table S1 shows the diagnostic performance of GB and RF for predicting binary outcomes, including Thal and IDA, in the training dataset. In this model, patients with IDA and Thal were not included in the training dataset because the small sample size might have affected the data analysis. The results demonstrated that both GB and RF achieved high accuracy in the training dataset (90.5% and 96.4%, respectively). The AUC-ROC values of GB and RF were 0.969 and 0.996, respectively. However, their performance slightly decreased in the testing dataset, with the accuracy decreasing to 90.7% (95% confidence interval [CI]: 86.8%–94.6%) for both GB and RF. Table 2 shows that their AUC-ROC remained consistent at 0.953 (95% CI: 0.924–0.982). Supplementary Table S2 shows the diagnostic performance of RF and GB for predicting multiclass outcomes, including Thal, IDA, and IDA with Thal, in the training dataset. RF achieved an accuracy of 91.7% compared with 85.4% for GB. The AUC-ROC values of GB and RF were 0.957 and 0.986, respectively. RF demonstrated a higher sensitivity than GB in predicting Thal and IDA groups. A notable decrease in sensitivity was observed both in RF and GB in predicting IDA with Thal, with RF achieving 87.0% sensitivity and GB achieving 78.0%. Table 3 shows the diagnostic performance of RF and GB for predicting multiclass outcomes, including Thal, IDA, and IDA with Thal, in the testing dataset. Both GB and RF exhibited lower accuracies in the testing dataset (80.4% [95% CI: 75.2%–85.2%] accuracy for GB and 82.2% [95% CI: 77.0%–87.0%] for RF) than in the training dataset. The AUC-ROC of both algorithms is also slightly lower than in the training dataset (0.910 [95% CI: 0.859–0.949] for GB and 0.899 [95% CI: 0.844–0.939] for RF). However, both algorithms maintained high sensitivity for predicting Thal (89.2% [95% CI: 83.3%–93.9%] for RF and 81.9% [95% CI: 74.8%–88.6%] for GB). By contrast, the sensitivity (76.8% [95% CI: 66.7%–86.8%]) was lower in the IDA group using RF. The sensitivity for predicting IDA with Thal was particularly low (69.1% [95% CI: 50.0%–86.4%] for GB and 65.3% [95% CI: 46.4%–84.6%] for RF). The two essential variables for predicting multiclass outcomes using GB and RF were MCHC and MCV (Figure 2). Table 4 presents the diagnostic performance of 15 previously reported formulas for predicting binary outcomes (Thal and IDA). Among these, only the Hct/Hb index demonstrated strong predictive capability, and it achieved an AUC-ROC of 0.820. Furthermore, our study revealed that the GB and RF algorithms, when utilizing only CBC indices, exhibited significantly higher predictive efficiency than any single index ( P < 0.05). Discussion IDA and Thal are common causes of hypochromic microcytic anemia in Southeast Asia, particularly in Thailand [30]. The differential diagnosis of these abnormalities is vital for effective treatment and proper genetic counseling. Serum ferritin and transferrin saturation are widely used for diagnosing IDA, but their accuracy can be influenced by various confounding factors. For example, serum ferritin thresholds must be adjusted in patients with concurrent inflammation. Moreover, conventional cut-offs for younger adults may not be suitable for older adults because of the cumulative effects of inflammation with age. To enhance accuracy and validity, serum ferritin cut-offs should be adjusted to demographic and physiological factors [31]. Our study used serum ferritin <30 ng/mL and transferrin saturation <16% as cut-off values because we included only adult patients without underlying conditions such as inflammation or pregnancy. Several RBC index formulas have been constructed to discriminate between IDA and TT. However, each formula has a different efficiency depending on the study population [19, 32, 33]. Applying discriminating formulas and indices for TT, TI, and IDA offers limited diagnostic value. Thus, the current study included both TT and TI in the Thal group, which generally occurs in real-world hospital situations. Our internal validation showed that both the RF and GB models performed well in discriminating IDA from Thal in either the training or testing datasets but not in the differential diagnosis of IDA with Thal. Thus, a patient’s history review, including data regarding the family history, blood transfusion, history of anemia, blood loss, melena, and hematochezia, might help conduct a proper investigation for the differential diagnosis of IDA with Thal. In this study, we utilized only RBC indices and personal demographic data to minimize feature redundancy and enhance the performance of our machine learning model. We demonstrated that MCHC and MCV levels are the two important features for machine learning in the RF and GB models, respectively. MCHC represents the average Hb concentration within a single RBC. Notably, this index is significantly lower in patients with IDA than in patients with Thal (TT and TI) (Table 1). This may be explained by the fact that IDA results from a lack of iron, which is essential for Hb production. As a result, RBCs have lower Hb content and are smaller in size. By contrast, Thal is caused by a defect in globin chain production, with iron supply remaining sufficient. Consequently, MCHC is not as drastically reduced in Thal as in IDA. A recent study used GB to predict individuals with TT, IDA, and a normal condition and identified MCV, MCH, RDW-SD, and Hb levels as the most significant features [34]. MCV and MCH are effective markers for screening Thal carriers [35-36]. Similarly, in the current study, MCV also emerged as a key feature in the model, thus aligning with previous findings. Additionally, we compared the diagnostic performance of previously reported formulas with machine learning models (GB and RF) in binary outcomes model. Among the 15 formulas used to predict Thal (TT and TI) and IDA, only the Hct/Hb index demonstrated strong performance. This is the first study to use the Hct/Hb index to discriminate between IDA and Thal (TT and TI). This index is useful for differential diagnosis because patients with Thal can have low Hb levels [37]. By contrast, patients with IDA have low RBC production, which contributes to low Hb and Hct levels [38]. However, the Hct/Hb index has not been used to differentiate IDA from TT in previous studies. Moreover, the performance of this single index remains lower than that of our machine learning model that uses only RBC indices (Table 4). The remaining 14 formulas demonstrated low performance in our cases because they were originally validated for predicting TT and IDA. However, the Thal group in our study included a diverse range of genotypes that encompasses both TT and TI. Therefore, the applicability of these formulas may be limited in regions where Thal is prevalent. We demonstrated the efficiency of two predictive models: one distinguishing between Thal and IDA and the other differentiating among Thal, IDA, and the combined group of IDA with Thal by using GB and RF. GB performed better in multiclass outcomes model, with an AUC-ROC of 0.910 (95% CI: 0.859–0.949). We suggested the application of a predictive model involving three groups for patient care because we could not exclude patients with both IDA and Thal in a real-world clinical setting. However, the accuracy of GB in the testing dataset was lower than that in the training dataset, and this result might have been due to the small sample size. The model performance of GB in the testing dataset remained high and acceptable. However, further prospective studies should be performed to externally validate the performance and refinement of this diagnostic model. Finally, a machine learning approach for discriminating between IDA and Thal using the GB algorithm was developed, along with a web-based prediction tool named “PSU Thal-IDA Pred.” The probability scores can guide clinicians in selecting suitable confirmation tests in first-visit patients with unknown causes of hypochromic microcytic anemia, resulting in reduced laboratory investigation costs and time and reduced blood volume in the sample collection of patients with anemia. Users can easily access our website at https://srisintornw.shinyapps.io/small_mcv_prediction_v2/. The prediction scores were obtained after inputting the RBC indices. In conclusion, our study demonstrated that machine learning (GB and RF) algorithms are efficient in discriminating between patients with IDA and Thal but not in complex diseases, such as IDA with Thal. Thus, we recommend the application of this diagnostic model for the diagnosis of IDA and Thal in the Thai population, wherein IDA and Thal are endemic. Declarations Acknowledgments This study was supported funding by the Faculty of Medicine, Prince of Songkla University (REC 62-232-5-2). Author contributions Conceptualization: WT, WS, and PS; Methodology, formal analysis, and investigation: WT, WS, TP, and PS; Writing–original draft preparation: WT and WS; Writing–review and editing: WT, WS, TP, and PS; Funding acquisition: WT. Data availability The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.Some data may not be available because of privacy or ethical reasons. Competing interests The authors declare no competing interests. Ethical approval The study was approved by the office of Human Research Ethics Committee, Faculty of Medicine, Prince of Songkla University, and was conducted in accordance with the Declaration of Helsinki (approval number: REC 62-232-5-2). The requirement for inform consent was waived because we used deidentified data. References Newhall, D. A., Oliver, R. & Lugthart, S. Anaemia: A disease or symptom. Neth . J . Med . 78 , 104-110 (2020). GBD 2021 Anaemia Collaborators. Prevalence, years lived with disability, and trends in anaemia burden by severity and cause, 1990-2021: findings from the Global Burden of Disease Study 2021. Lancet Haematol . 10 , e713-e734 (2023). Winichagoon, P. Prevention and control of anemia: Thailand experiences. J . Nutr . 132 (Supplement), 862S-866S (2002). Sirachainan, N. et al . New mathematical formula for differentiating thalassemia trait and iron deficiency anemia in thalassemia prevalent area: a study in healthy school-age children. Southeast Asian J . Trop . Med . Public Health 45 , 174-182 (2014). Fucharoen, S. & Winichagoon, P. Haemoglobinopathies in Southeast Asia. Indian J . Med . Res . 134 , 498-506 (2011). Johnson-Wimbley, T. D. & Graham, D. Y. Diagnosis and management of iron deficiency anemia in the 21st century. Ther . Adv . Gastroenterol . 4 , 177-184 (2011). Nopparatana, C., Nopparatana, C., Saechan, V., Karnchanaopas, S. & Srewaradachpisal, K. Prenatal diagnosis of α- and β-thalassemias in southern Thailand. Int . J . Hematol . 111 , 284-292 (2020). Pornprasert, S., Thongsat, C. & Panyachadporn, U. Evaluation of applying a combination of red cell indexes and formulas to differentiate β-thalassemia trait from iron deficiency anemia in the Thai population. Hemoglobin 41 , 116-119 (2017). Ambayya, A. et al . Haematological reference intervals in a multiethnic population. PLOS One 9 , e91968 (2014). Huang, T. C. et al . Discrimination index of microcytic anemia in young soldiers: a single institutional analysis. PLOS One 10 , e0114061 (2015). Laengsri, V. et al . ThalPred: a web-based prediction tool for discriminating thalassemia trait and iron deficiency anemia. BMC Med . Inform . Decis . Mak . 19 , 212 (2019). Hasani, M. & Hanani, A. Automated diagnosis of iron deficiency anemia and thalassemia by data mining techniques. IJCSNS 17 , 326-331 (2017). Çil, B., Ayyıldız, H. & Tuncer, T. Discrimination of β-thalassemia and iron deficiency anemia through extreme learning machine and regularized extreme learning machine based decision support system. Med . Hypotheses 138 , 109611 (2020). Karabayir, I., Goldman, S. M., Pappu, S. & Akbilgic, O. Gradient boosting for Parkinson's disease diagnosis from voice recordings. BMC Med . Inform . Decis . Mak . 20 , 228 (2020). Camaschella, C. Iron-deficiency anemia. N . Engl . J . Med . 372 , 1832-1843 (2015). Chong, S. S., Boehm, C. D., Higgs, D. R. & Cutting, G. R. Single-tube multiplex-PCR screen for common deletional determinants of alpha-thalassemia. Blood 95 , 360-362 (2000). James, G., Witten, D., Hastie, T. & Tibshirani, R. An Introduction to Statistical Learning : with Applications in R (Springer US, New York, 2021).https://doi.org/10.32614/CRAN.package.ISLR2. Chawla, N. V., Bowyer, K. W., Hall, L. O. & Kegelmeyer, W. P. Smote: synthetic minority over-sampling technique. J . Artifical Intell . Res . 16 , 321-357 (2002). Pornprasert, S., Panya, A., Punyamung, M., Yanola, J. & Kongpan, C. Red cell indices and formulas used in differentiation of β-thalassemia trait from iron deficiency in Thai school children. Hemoglobin 38 , 258-261 (2014). Nalbantoğlu, B. et al . Indices used in differentiation of thalassemia trait from iron deficiency anemia in pediatric population: are they reliable? Pediatr . Hematol . Oncol . 29 , 472-478 (2012). Demir, A., Yarali, N., Fisgin, T., Duru, F. & Kara, A. Most reliable indices in differentiation between thalassemia trait and iron deficiency anemia. Pediatr . Int . 44 , 612-616 (2002). Sirdah, M., Tarazi, I., Al Najjar, E. & Al Haddad, R. Evaluation of the diagnostic reliability of different RBC indices and formulas in the differentiation of the beta-thalassaemia minor from iron deficiency in Palestinian population. Int . J . Lab . Hematol . 30 , 324-330 (2008). DeLong, ER., DeLong, DM. & Clarke-Pearson, DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 44, 837-845 (1988). R : A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, Vienna, Austria, 2022). https://www.R-project.org. Kuhn, M. & Wickham, H. Tidymodels : a Collection of Packages for Modeling and Machine Learning Using Tidyverse Principles (2020). https://www.tidymodels.org. Wright, M.N. & Ziegler, A. ranger: A Fast Implementation of Random Forests for High Dimensional Data in C++ and R. J . Stat . Softw . 77 , 1-17 (2017). Chen, T. et al . Xgboost: extreme Gradient Boosting. (2022); 10.32614/CRAN.package.xgboost. Themis, H. E. Extra Recipes Steps for Dealing with Unbalanced Data. R package. version 1.0.3. (2025). 10.32614/CRAN.package.themis. Robin, X. et al . pROC: an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics . 12 , 77 (2011). Pansuwan, A., Fucharoen, G., Fucharoen, S., Himakhun, B. & Dangwiboon, S. Anemia, iron deficiency and thalassemia among adolescents in Northeast Thailand: results from two independent surveys. Acta Haematol . 125 , 186-192 (2011). Naveed K, et al . Defining ferritin clinical decision limits to improve diagnosis and treatment of iron deficiency: A modified Delphi study. Int . J . Lab . Hematol . 45 , 377-386 (2023). Carla, M. G., Rafael, S. P., Isabel, F. G., Cristina, G. F. & Teresa, S. M. New haematologic score to discriminate beta thalassemia trait from iron deficiency anaemia in a Spanish Mediterranean region. Clin . Chim . Acta . 507 , 69-74 (2020). Zaghloul, A. et al . Introduction of new formulas and evaluation of the previous red blood cell indices and formulas in the differentiation between beta thalassemia trait and iron deficiency anemia in the Makkah region. Hematology 21 , 351-358 (2016). Wang, W., Ye, R., Tang, B. & Qi, Y. MultiThal-classifier, a machine learning-based multi-class model for thalassemia diagnosis and classification. Clin . Chim . Acta 567 , 120025 (2025). Yamsri, S. et al . Prevention of severe thalassemia in northeast Thailand: 16 years of experience at a single university center. Prenat Diagn . 30 ,540-546 (2010). Chaitraiphop, C. et al . Thalassemia Screening Using Different Automated Blood Cell Counters: Consideration of Appropriate Cutoff Values. Clin Lab . 62 , 545-52 (2016). Weatherall, D. J. & Clegg, J. B. The Thalassemia Syndromes . 4th ed (Blackwell Science., Oxford, 2001). Lopez, A., Cacoub, P., Macdougall, I. C. & Peyrin-Biroulet, L. Iron deficiency anaemia. Lancet 387 , 907-916 (2016). Tables Table 1 . Baseline characteristics and hematological features of patients with Thal and IDA. Characteristic Thal (n = 635) (mean ± SD) IDA with Thal (n = 126) (mean ± SD) IDA (n = 382) (mean ± SD) P -value 1 Sex, female (n, %) 428 (67.4%) 111 (88.1%) 327 (85.6%) <0.001 Age (years) 52 ± 21 41 ± 15 47 ± 19 <0.001 Red blood cells (×10 6 /µL) 4.56 ± 0.9 4.56 ± 0.8 4.29 ± 0.7 <0.001 Hemoglobin (g/dL) 9.64 ± 1.9 8.67 ± 2.2 8.65 ± 1.9 <0.001 Hematocrit (%) 30.5 ± 5.9 28.5 ± 6.0 29.3 ± 5.4 <0.001 Mean cell volume (fL) 67.0 ± 8.0 63.0 ± 9.0 68.0 ± 7.0 <0.001 Mean corpuscular hemoglobin (pg) 21.4 ± 2.9 19.0 ± 3.7 20.1 ± 3.2 <0.001 Mean corpuscular hemoglobin concentration (g/dL) 31.65 ± 2.0 30.2 ± 2.3 29.3 ± 2.0 <0.001 Red blood cell distribution width (%) 18.6 ± 4.9 20.0 ± 4.1 19.0 ± 3.4 <0.001 Thalassemia type (n) α-Thal trait 9 6 0 Hb Constant Spring trait 18 4 0 Hb H disease 81 2 0 Hb H with Constant Spring 19 0 0 Hb H with Hb E trait 1 0 0 β-Thal trait 194 26 0 β + /β + -Thal disease 9 0 0 β + -Thal/Hb E disease 21 1 0 Hb E trait 218 80 0 Hb E trait with Hb Constant Spring trait 2 0 0 Homozygous Hb E 51 5 0 HPFH trait 12 2 0 Thal, thalassemia; IDA, iron deficiency anemia; HPFH, hereditary persistence of fetal hemoglobin. 1 : Pearson’s chi-squared test; Kruskal–Wallis rank sum test. Table 2 . Diagnostic performance of the RF and GB models for predicting binary outcomes (Thal [TT and TI] and IDA) in the testing dataset. Metric GB RF Median 95% CI Median 95% CI Sensitivity (%) 93.8 89.4–97.7 93.9 89.5–97.7 Specificity (%) 85.7 78.0–92.8 85.9 77.6–93.1 Accuracy (%) 90.7 86.8–94.6 90.7 86.8–94.6 Kappa 0.802 0.719–0.881 0.803 0.717–0.887 AUC-ROC 0.953 0.924–0.982 0.953 0.924–0.982 TT, thalassemia trait; TI, Thal intermedia; IDA, iron deficiency anemia; RF, random forest; GB, gradient boosting; AUC-ROC, area under the receiver operating characteristic curve. Table 3 . Diagnostic performance of the RF and GB models for predicting multiclass outcomes (Thal [TT and TI], IDA, and IDA with Thal) in the testing dataset. Metric GB RF Thal IDA IDA with Thal Thal IDA IDA with Thal Sensitivity, % (95% CI) 81.9 (74.8–88.6) 82.1 (73.0–90.4) 69.1 (50.0–86.4) 89.2 (83.3–93.9) 76.8 (66.7–86.8) 65.3 (46.4–84.6) Specificity, % (95% CI) 88.5 (82.1–94.3) 91.7 (86.8–95.4 90.3 (86.1–94.0) 84.5 (77.0–90.0) 93.6 (89.6–96.9) 92.6 (88.9–96.1) Accuracy, % (95% CI) 80.4 (75.2–85.2) 82.2 (77.0–87.0) Kappa, median (95% CI) 0.669 (0.586–0.751) 0.689 (0.603–0.770) AUC-ROC, median (95% CI) 0.910 (0.859–0.949) 0.899 (0.844–0.939) Thal, thalassemia; TT, Thal trait; TI, Thal intermedia; IDA, iron deficiency anemia; RF, random forest; GB, gradient boosting; AUC-ROC, area under the receiver operating characteristic curve. Table 4 . Diagnostic performance of a single formula and comparison of AUC-ROC between each formula with GB and RF algorithms. Formula Cut-off value 95% CI Sensitivity (%) 95% CI Specificity (%) 95% CI Youden’s index 95% CI AUC-ROC of single formula AUC-ROC of GB AUC-ROC of RF P -value 1 P -value 2 Mentzer 16.4 13.0–19.7 93.6 93.3–94.0 31.8 27.5–36.2 25.5 21.5–29.6 0.568 0.953 0.953 8.65E-22 1.55E-23 Shine & Lal 1221.0 942.5–1499.5 78.4 62.4–94.5 32.0 14.5–49.5 10.5 10.0–11.9 0.529 0.953 0.953 1.65E-19 1.57E-21 England and Fraser 9.1 7.3–10.8 88.9 82.0–95.8 44.0 34.6–53.4 32.9 30.4–35.4 0.640 0.953 0.953 7.36E-20 1.01E-18 Srivastava 5.1 4.6–5.6 72.5 60.3–84.7 41.4 34.5–48.4 14.0 8.8–19.2 0.478 0.953 0.953 2.40E-31 3.16E-30 Green and King 82.7 82.5–82.9 86.7 82.6–90.7 53.0 49.5–56.5 39.6 32.0–47.2 0.664 0.953 0.953 3.96E-16 6.33E-15 Jayabose 240.0 232.8–247.2 92.7 92.5–92.9 35.4 30.8–39.9 28.1 23.7–32.5 0.596 0.953 0.953 4.27E-20 5.86E-21 Ricera 3.5 3.1–3.9 82.0 68.9–95.1 41.9 25.0–58.8 23.9 20.0–27.7 0.559 0.953 0.953 1.63E-24 7.07E-24 Ehsani 21.1 14.5–27.6 70.9 53.3–88.4 54.7 36.7–72.8 25.6 25.0–26.2 0.580 0.953 0.953 2.44E-21 7.39E-21 Sirdah 33.0 29.2–36.9 81.6 65.3–97.8 48.8 35.1–62.5 30.4 27.8–33.0 0.643 0.953 0.953 4.64E-18 2.27E-18 Kerman_i 381.0 329.7–432.2 79.0 65.2–92.7 35.9 20.3–51.5 14.9 13.0–16.7 0.512 0.953 0.953 7.90E-25 1.16E-24 Kerman_ii 94.7 75.9–113.5 73.9 55.2–92.5 51.2 36.2–66.1 25.1 21.4–28.7 0.579 0.953 0.953 2.45E-19 1.10E-20 Keikhaei 24.4 24.4–24.4 92.8 92.4–93.2 43.2 41.7–44.7 36.0 34.1–37.9 0.596 0.953 0.953 2.74E-20 7.02E-21 Matos 22.8 22.4–23.1 85.6 82.4–88.8 50.0 40.2–59.8 35.6 29.1–42.1 0.686 0.953 0.953 1.27E-16 1.27E-16 MCV/Hb 6.5 6.5–6.5 89.8 88.9–90.7 42.6 38.1–47.1 32.4 27.0–37.7 0.635 0.953 0.953 3.85E-20 4.39E-20 Hct/Hb 3.2 3.2–3.3 77.9 65.0–90.8 73.8 61.5–86.1 51.7 51.0–52.4 0.820 0.953 0.953 1.12E-07 1.49E-07 1 P -value of AUC-ROC comparison between single formula and GB, 2 P -value of AUC-ROC comparison between single formula and RF. GB, gradient boosting; RF, random forest; AUC-ROC, area under the receiver operating characteristic curve. Additional Declarations No competing interests reported. Supplementary Files SupplementaryTableS1S2.docx Cite Share Download PDF Status: Published Journal Publication published 15 May, 2025 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 14 Apr, 2025 Reviews received at journal 13 Apr, 2025 Reviews received at journal 29 Mar, 2025 Reviewers agreed at journal 21 Mar, 2025 Reviewers agreed at journal 21 Mar, 2025 Reviewers invited by journal 20 Mar, 2025 Submission checks completed at journal 20 Mar, 2025 First submitted to journal 20 Mar, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5623304","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":431924781,"identity":"f5ab92ab-79cb-419d-85f1-1228476bc522","order_by":0,"name":"Wanicha Tepakhan","email":"","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":false,"prefix":"","firstName":"Wanicha","middleName":"","lastName":"Tepakhan","suffix":""},{"id":431924782,"identity":"180f6673-0284-4839-a8c7-a9ecac000254","order_by":1,"name":"Wisarut Srisintorn","email":"","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":false,"prefix":"","firstName":"Wisarut","middleName":"","lastName":"Srisintorn","suffix":""},{"id":431924783,"identity":"e84c0537-5fb7-40ec-9ea8-50b61a085c67","order_by":2,"name":"Tipparat Penglong","email":"","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":false,"prefix":"","firstName":"Tipparat","middleName":"","lastName":"Penglong","suffix":""},{"id":431924784,"identity":"92f88278-3d9d-40bf-b34e-7470b430c109","order_by":3,"name":"Pirun Saelue","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3ElEQVRIiWNgGAWjYDACZhBRIcHDz94D5vPwEafljIWMZM8ZBoYDQC1sRNnE2FZhY3AjB6yFgaAWc3beZ9IFbBI8kjPfHnz8McdOho2B+eGjG3i0WDazm0nP4AH6RTov2eDgtmSgw9iMjXPwaDE4zMYmzSMBtGV2jpnEwW3MQC08bNKEtRhI8BjcPAPSUk+slgSglhs8IC2HCWuxbGZjtuY5AHRYT46xwdltx3nYmAn4xZz/GONt3n919vzsZwwfVG6rBjKaHz7G6zBMIWY8ynFoGQWjYBSMglGABgCLtjn9JsEVQQAAAABJRU5ErkJggg==","orcid":"","institution":"Prince of Songkla University","correspondingAuthor":true,"prefix":"","firstName":"Pirun","middleName":"","lastName":"Saelue","suffix":""}],"badges":[],"createdAt":"2024-12-11 10:23:06","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5623304/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5623304/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-025-01458-5","type":"published","date":"2025-05-15T15:56:59+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":79177047,"identity":"47e2a484-c59e-4763-b35c-c217e67e9e8c","added_by":"auto","created_at":"2025-03-25 10:04:06","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":140299,"visible":true,"origin":"","legend":"\u003cp\u003eStudy population flow and distribution of cases in the training and testing dataset.\u003c/p\u003e\n\u003cp\u003eCBC, complete blood count; MCV, mean corpuscular volume\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5623304/v1/1a20d0545b317a2571d9241b.png"},{"id":79179186,"identity":"1e9d9228-7a1a-4381-808c-00a8a831ecf2","added_by":"auto","created_at":"2025-03-25 10:12:06","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":39370,"visible":true,"origin":"","legend":"\u003cp\u003eVariable importance from multiclass outcomes (Thal, IDA, IDA with Thal) of the random forest model \u003cstrong\u003e(a)\u003c/strong\u003eand gradient boosting model \u003cstrong\u003e(b).\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5623304/v1/009ca5ec3f20bc65051c1ca3.png"},{"id":83067895,"identity":"67e149bd-3a93-4a2e-835d-a1ac98ea5e49","added_by":"auto","created_at":"2025-05-19 16:07:46","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":913354,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5623304/v1/d2ebc8dc-3fc0-42f6-9bde-963f2825edf1.pdf"},{"id":79177048,"identity":"4cfb2ca2-ec0b-446a-8529-4a800f9f10af","added_by":"auto","created_at":"2025-03-25 10:04:06","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":16691,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1S2.docx","url":"https://assets-eu.researchsquare.com/files/rs-5623304/v1/ca96fe105c10630a547a95e5.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Machine learning approach for differentiating iron deficiency anemia and thalassemia using random forest and gradient boosting algorithms","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAnemia is a common condition encountered in clinical practice. It is defined as a low number of red blood cells (RBCs) or a low hemoglobin (Hb) concentration. Anemia is classified into three categories on the basis of the Hb concentration and RBC size: hypochromic microcytic anemia, normochromic normocytic anemia, and macrocytic anemia [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. Iron deficiency anemia (IDA) and thalassemia (Thal) are the most common causes of hypochromic microcytic anemia. A 2021 global survey reported that the prevalence of anemia was 24.3%, and approximately 66.2% of the total anemia cases are caused by IDA [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. IDA is characterized by a depleted iron storage that leads to low RBC production. Moreover, it can be caused by a low iron intake, acute or chronic blood loss, or abnormalities in iron absorption. The prevalence of IDA is approximately 1.5\u0026ndash;12% in the Thai population [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Thal is one of the most common causes of anemia [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. It is an inherited disorder caused by a mutation in the globin gene that results in reduced or absent globin chain production. Patients with Thal traits (TTs) usually exhibit no anemic symptoms. By contrast, patients with Thal disease show widely different clinical phenotypes (from mild to severe anemia) depending on the mutation type. In Thailand, the prevalence rates of TTs, including α-TT, β-TT, and heterozygous Hb E, are approximately 20\u0026ndash;30%, 3\u0026ndash;9%, and 10\u0026ndash;50%, respectively [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eLaboratory investigations for diagnosing these conditions include serum iron tests, ferritin level assessment, Hb analysis, and deoxyribonucleic acid (DNA) analysis for Thal [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. However, Hb and DNA analyses are unavailable in some hospitals owing to the need for specialized equipment, advanced technical expertise, and the time-consuming nature of the tests. In addition, these investigations can be costly, as patients often incur substantial expenses when physicians request comprehensive confirmatory tests for the diagnosis. Therefore, several mathematical formulas based on RBC indices, including Sirdah, Green and King, Mentzer, England and Fraser, Ehsani, Srivastava, Shine \u0026amp; Lal, and the 11T score, have been developed to help clinicians select appropriate confirmatory tests for differentiating between IDA and TTs, aiming to reduce investigation costs and time [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. However, the efficiency of these equations varies. Moreover, the cut-off values of each formula are affected by sex, age, and ethnicity, resulting in unsatisfactory sensitivity and specificity results among different populations [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. Applying these formulas for the differential diagnosis between IDA and Thal, including TT and Thal intermedia (TI), offers limited diagnostic value.\u003c/p\u003e \u003cp\u003eIn recent years, several machine learning algorithms such as C4.5 decision tree, k-nearest neighbor, artificial neural network, support vector machine, Naive Bayes, random forest (RF), vote algorithm, and extreme learning machine were evaluated for their classification performance in predicting whether a patient has IDA or TT [\u003cspan additionalcitationids=\"CR12\" citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. RF showed a high performance with accuracies of 94.17% and 96.0% for IDA and TT, respectively [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. In addition, the gradient boosting (GB) algorithm has been an effective model for predicting and diagnosing several diseases [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]. However, this algorithm has limited information in discriminating between patients with IDA and Thal.\u003c/p\u003e \u003cp\u003eThus, owing to the advancements of machine learning algorithms and the limitations of previous formulas for the differential diagnosis of IDA and Thal, this study aimed to generate a diagnostic model by using RF and GB algorithms to predict the probability of IDA and Thal. The results of the study should aid clinicians in determining appropriate laboratory investigations for patients with hypochromic microcytic anemia.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003e\u003cem\u003eParticipants\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis cross-sectional study was conducted at Songklanagarind Hospital, the largest tertiary hospital in southern Thailand, between January 2015 and December 2019.\u0026nbsp;We assessed the first-visit data of 7,488 patients, who had the following characteristics:\u0026nbsp;1)\u0026nbsp;age \u0026gt;15 years, 2)\u0026nbsp;Hb concentration \u0026lt;13 g/dL\u0026nbsp;in men and menopausal women or \u0026lt;12\u0026nbsp;g/dL in reproductive women, 3)\u0026nbsp;mean corpuscular volume\u0026nbsp;(MCV)\u0026nbsp;\u0026lt;80 fL, 4)\u0026nbsp;available\u0026nbsp;iron profiles and ferritin level data, and 5)\u0026nbsp;Hb and DNA analyses for Thal.\u0026nbsp;Patients with anemia of inflammation, transfusion-dependent Thal, pregnancy, or incomplete laboratory data were excluded.\u0026nbsp;To exclude anemia due to inflammation\u0026nbsp;and pregnancy, a hematologist reviewed the medical records to confirm the diagnoses of IDA and Thal and to exclude patients\u0026nbsp;with inflammation and infection.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePatients with serum ferritin levels \u0026lt;30 ng/mL and transferrin saturation\u0026nbsp;\u0026lt;16%\u0026nbsp;were diagnosed with IDA\u0026nbsp;[15].\u0026nbsp;All patients were diagnosed with Thal\u0026nbsp;(TT and TI)\u0026nbsp;by\u0026nbsp;using the following diagnostic criteria:\u0026nbsp;patients with Hb type A2A and Hb A2 levels\u0026nbsp;≥3.5%\u0026nbsp;were diagnosed with β-TT.\u0026nbsp;Those with Hb type A2A, Hb A2 levels \u0026lt;3.5%,\u0026nbsp;and positive α-Thal mutation following DNA analysis were diagnosed with α-TT.\u0026nbsp;Those with Hb type EA and Hb E \u0026gt;10%–35%\u0026nbsp;were considered\u0026nbsp;to have the Hb\u0026nbsp;E trait.\u0026nbsp;Patients\u0026nbsp;diagnosed with TI exhibited Hb patterns such as A2FA, EFA, EE, A2AH, A2ABart'sH, CSA2AH, CSA2ABart'sH, EABart's, EFABart's, CSEABart's, and CSEFABart's; furthermore, these patients had no\u0026nbsp;history\u0026nbsp;of transfusion, and their conditions\u0026nbsp;were\u0026nbsp;confirmed through DNA analysis.\u0026nbsp;The\u0026nbsp;definitions and full names of the abbreviations are shown in the appendix.\u0026nbsp;Patients who met the criteria for IDA and Thal were diagnosed as having IDA with\u0026nbsp;Thal.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eEthical approval\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eEthical approval for this study was obtained from the Human Research Ethics Committee\u0026nbsp;(HREC)\u0026nbsp;of the Faculty of Medicine, Prince of Songkla University\u0026nbsp;(REC 62-232-5-2).\u0026nbsp;The HREC waived the requirement for informed consent because this study used deidentified data.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eLaboratory techniques\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe hematological features were measured using an automated blood cell counter\u0026nbsp;(XN3000; Sysmex Corp., Kobe, Japan).\u0026nbsp;Hb analysis was performed using capillary electrophoresis\u0026nbsp;(CapillaryS2; Sebia, Lisses, France).\u0026nbsp;Serum\u0026nbsp;iron levels, total iron binding capacity, and ferritin levels were measured using an automated analyzer\u0026nbsp;(Cobas e411; Roche, Rotkreuz, Switzerland).\u0026nbsp;DNA analysis for Thal was performed using polymerase chain reaction and reverse dot blot hybridization, as previously described\u0026nbsp;[7, 16].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u003cem\u003eStatistical analysis\u003c/em\u003e\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe baseline characteristics and hematological features of patients with Thal and IDA were compared using Pearson’s chi-squared test for categorical data\u0026nbsp;and the Kruskal–Wallis rank sum test for continuous data.\u0026nbsp;\u003cem\u003eP\u003c/em\u003e \u0026lt;\u0026nbsp;0.05 was\u0026nbsp;considered statistically significant.\u0026nbsp;The complete blood count\u0026nbsp;(CBC)\u0026nbsp;data of patients diagnosed with Thal\u0026nbsp;(TT and TI), IDA, and IDA with Thal\u0026nbsp;(TT and TI)\u0026nbsp;were divided into training and testing sets by using a ratio of 80:20.\u0026nbsp;Nine\u0026nbsp;features were used in the machine learning methods, including Hb levels, hematocrit\u0026nbsp;(Hct), MCV, mean corpuscular hemoglobin\u0026nbsp;(MCH), mean corpuscular hemoglobin concentration\u0026nbsp;(MCHC),\u0026nbsp;red cell distribution width\u0026nbsp;(RDW), RBC count, age, and sex.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTwo types of models were built:\u0026nbsp;binary outcome models\u0026nbsp;(Thal and IDA)\u0026nbsp;and multiclass outcome models\u0026nbsp;(Thal, IDA, and IDA with Thal).\u0026nbsp;Two ensemble machine learning classification methods were used to diagnose Thal and/or IDA.RF builds multiple decision trees on random subsets of the training dataset.\u0026nbsp;This reduces the correlation between trees and avoids overfitting by selecting random subsets of features at each split.\u0026nbsp;The final prediction is the majority vote from all\u0026nbsp;the trees.\u0026nbsp;GB builds trees sequentially\u0026nbsp;by\u0026nbsp;using information learned from previous trees.\u0026nbsp;In theory, this improves accuracy compared to\u0026nbsp;single-decision trees\u0026nbsp;[17].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe model features were optimized using 10-fold cross-validation to maximize the area under the receiver operating characteristic curve\u0026nbsp;(AUC-ROC).\u0026nbsp;The Latin hypercube method was used to sample 1,000 sets of parameter values for each model.\u0026nbsp;The best features were those that maximized the AUC-ROC\u0026nbsp;of the validation data.\u0026nbsp;The performance on the test set was evaluated using a range of metrics, including accuracy, kappa coefficient, sensitivity, specificity and AUC-ROC.\u0026nbsp;To address the class imbalance, the synthetic minority over-sampling technique\u0026nbsp;[18]\u0026nbsp;was employed to generate synthetic samples for the minority class, thus effectively balancing the dataset prior to model training.\u0026nbsp;A\u0026nbsp;comparison between the diagnostic performances of GB and RF with formulas based on RBC indices such as Hct/Hb, MCV/Hb, Keikhaei, Jayabose, Sirdah, Green and King, Mentzer, England and Fraser, Srivastava, Shine \u0026amp; Lal, Matos, Ricera, Kerman I, Kerman II, Ehsani\u0026nbsp;[8, 19-22]\u0026nbsp;was made\u0026nbsp;by using the method proposed by Delong et\u0026nbsp;al.\u0026nbsp;[23]. The definitions of all features and formulas are shown in the appendix.\u003c/p\u003e\n\u003cp\u003eAnalysis was performed using R version 4.4.2\u0026nbsp;[24].\u0026nbsp;The\u0026nbsp;model\u0026nbsp;specifications and analytical processes were performed\u0026nbsp;using \u003cem\u003etidymodels\u003c/em\u003e version 1.2\u0026nbsp;[25].\u0026nbsp;The underlying analytic packages for RF and GB were \u003cem\u003eranger\u003c/em\u003e\u003cem\u003eversion 0\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e17\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e0\u003c/em\u003e [26]\u0026nbsp;and \u003cem\u003exgboost\u003c/em\u003e\u003cem\u003eversion 1\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e7\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e8\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e1\u003c/em\u003e [27], respectively.\u0026nbsp;The over-sampling\u0026nbsp;of the minority class was performed\u0026nbsp;using\u0026nbsp;\u003cem\u003ethemis\u0026nbsp;\u003c/em\u003eversion 1.0.3 [28]. AUC-ROC was calculated using pROC version 1.18.5 [29].\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eA total of 14,407 CBC records of patients with anemia and low MCV from 2015 to 2019\u0026nbsp;were assessed.\u0026nbsp;First-visit data from 7,488 patients were selected.\u0026nbsp;In total, 6,345\u0026nbsp;patients with anemia of inflammation\u0026nbsp;(n\u0026nbsp;=\u0026nbsp;1,535),\u0026nbsp;transfusion-dependent Thal\u0026nbsp;(n\u0026nbsp;=\u0026nbsp;65),\u0026nbsp;or incomplete laboratory records\u0026nbsp;(n\u0026nbsp;=\u0026nbsp;4,745)\u0026nbsp;were excluded.\u0026nbsp;Therefore, 1,143 patients were included in the study.\u0026nbsp;The data were randomly divided into two sets, namely, the training and testing datasets, by using a ratio of 80:20.\u0026nbsp;Figure 1 shows\u0026nbsp;the\u0026nbsp;distribution of\u0026nbsp;the cases in the training and testing datasets.\u0026nbsp;Table 1 summarizes the\u0026nbsp;baseline characteristics of the study groups.\u0026nbsp;All RBC indices differed significantly among the three groups.\u003c/p\u003e\n\u003cp\u003eSupplementary Table S1 shows the\u0026nbsp;diagnostic performance of\u0026nbsp;GB and RF for predicting binary outcomes, including\u0026nbsp;Thal and IDA, in the training dataset.\u0026nbsp;In this model, patients with IDA and Thal were not included in the training dataset because the small sample size might have affected the data analysis.\u0026nbsp;The results demonstrated that both GB and RF achieved high accuracy in the training dataset\u0026nbsp;(90.5%\u0026nbsp;and 96.4%, respectively).\u0026nbsp;The AUC-ROC values of GB and RF were 0.969 and 0.996, respectively.\u0026nbsp;However, their performance slightly\u0026nbsp;decreased in the testing dataset, with\u0026nbsp;the accuracy decreasing to 90.7% (95%\u0026nbsp;confidence interval\u0026nbsp;[CI]:\u0026nbsp;86.8%\u0026ndash;94.6%)\u0026nbsp;for both GB and RF.\u0026nbsp;Table 2 shows that their AUC-ROC remained consistent at 0.953\u0026nbsp;(95%\u0026nbsp;CI:\u0026nbsp;0.924\u0026ndash;0.982).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSupplementary Table S2 shows the\u0026nbsp;diagnostic performance of RF and GB for predicting multiclass outcomes, including Thal, IDA, and IDA with Thal, in the training dataset.\u0026nbsp;RF achieved an accuracy of 91.7%\u0026nbsp;compared with 85.4%\u0026nbsp;for GB.\u0026nbsp;The AUC-ROC values of GB and RF were\u0026nbsp;0.957 and 0.986, respectively.\u0026nbsp;RF demonstrated\u0026nbsp;a higher sensitivity\u0026nbsp;than GB in predicting Thal and IDA groups.\u0026nbsp;A notable decrease in sensitivity was observed both\u0026nbsp;in\u0026nbsp;RF and GB in predicting IDA with\u0026nbsp;Thal, with RF achieving 87.0%\u0026nbsp;sensitivity and GB achieving 78.0%.\u003c/p\u003e\n\u003cp\u003eTable 3 shows the\u0026nbsp;diagnostic performance of RF and GB for predicting multiclass outcomes, including Thal, IDA, and IDA with Thal, in the testing dataset.\u0026nbsp;Both GB and RF exhibited lower accuracies in the testing dataset\u0026nbsp;(80.4% [95%\u0026nbsp;CI:\u0026nbsp;75.2%\u0026ndash;85.2%]\u0026nbsp;accuracy for GB and\u0026nbsp;82.2% [95%\u0026nbsp;CI:\u0026nbsp;77.0%\u0026ndash;87.0%]\u0026nbsp;for RF)\u0026nbsp;than in the training dataset.\u0026nbsp;The AUC-ROC of both algorithms\u0026nbsp;is also slightly lower than in the training dataset\u0026nbsp;(0.910\u0026nbsp;[95%\u0026nbsp;CI:\u0026nbsp;0.859\u0026ndash;0.949]\u0026nbsp;for GB and 0.899\u0026nbsp;[95%\u0026nbsp;CI:\u0026nbsp;0.844\u0026ndash;0.939]\u0026nbsp;for RF).\u0026nbsp;However, both algorithms maintained high sensitivity for predicting Thal\u0026nbsp;(89.2% [95%\u0026nbsp;CI:\u0026nbsp;83.3%\u0026ndash;93.9%]\u0026nbsp;for RF and 81.9% [95%\u0026nbsp;CI:\u0026nbsp;74.8%\u0026ndash;88.6%]\u0026nbsp;for GB).\u0026nbsp;By contrast,\u0026nbsp;the sensitivity\u0026nbsp;(76.8% [95%\u0026nbsp;CI:\u0026nbsp;66.7%\u0026ndash;86.8%])\u0026nbsp;was lower\u0026nbsp;in the IDA group\u0026nbsp;using RF.\u0026nbsp;The sensitivity for predicting IDA with\u0026nbsp;Thal\u0026nbsp;was particularly\u0026nbsp;low\u0026nbsp;(69.1% [95%\u0026nbsp;CI:\u0026nbsp;50.0%\u0026ndash;86.4%]\u0026nbsp;for GB\u0026nbsp;and\u0026nbsp;65.3% [95%\u0026nbsp;CI:\u0026nbsp;46.4%\u0026ndash;84.6%]\u0026nbsp;for RF).\u0026nbsp;The two\u0026nbsp;essential variables for predicting multiclass outcomes using GB and RF were MCHC and MCV\u0026nbsp;(Figure 2).\u003c/p\u003e\n\u003cp\u003eTable 4 presents the diagnostic performance of 15 previously reported formulas for predicting binary outcomes\u0026nbsp;(Thal and IDA).\u0026nbsp;Among these, only the Hct/Hb index demonstrated strong predictive capability, and it achieved an AUC-ROC of 0.820.\u0026nbsp;Furthermore, our study revealed that the GB and RF algorithms, when utilizing only CBC indices, exhibited significantly higher predictive efficiency than any single index\u0026nbsp;(\u003cem\u003eP\u0026nbsp;\u003c/em\u003e\u0026lt; 0.05).\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIDA and Thal are common causes of hypochromic microcytic anemia in Southeast Asia, particularly in Thailand\u0026nbsp;[30].\u0026nbsp;The differential diagnosis of these abnormalities is vital for effective treatment and proper genetic counseling. \u0026nbsp;Serum ferritin and transferrin saturation are widely used for diagnosing IDA, but their accuracy can be influenced by various confounding factors.\u0026nbsp;For example, serum ferritin thresholds must be adjusted in patients with concurrent inflammation.\u0026nbsp;Moreover, conventional cut-offs for younger adults may not be suitable for older adults because of the cumulative effects of inflammation with age.\u0026nbsp;To enhance accuracy and validity, serum ferritin cut-offs should be\u0026nbsp;adjusted\u0026nbsp;to demographic and physiological factors\u0026nbsp;[31].\u0026nbsp;Our study used serum ferritin \u0026lt;30 ng/mL and transferrin saturation \u0026lt;16%\u0026nbsp;as cut-off values because we included only adult patients without underlying conditions such as inflammation or pregnancy.\u0026nbsp;Several RBC index formulas have been constructed to discriminate between IDA and TT.\u0026nbsp;However, each formula has a different efficiency depending on the study population\u0026nbsp;[19, 32, 33].\u0026nbsp;Applying\u0026nbsp;discriminating formulas and indices for TT, TI, and IDA offers\u0026nbsp;limited diagnostic value.\u0026nbsp;Thus, the current study included both TT and TI in the Thal group, which generally occurs in real-world\u0026nbsp;hospital situations.\u0026nbsp;Our internal validation showed that both\u0026nbsp;the RF and GB models performed well in discriminating IDA from Thal in either the training or testing datasets but not in the differential diagnosis of\u0026nbsp;IDA with\u0026nbsp;Thal.\u0026nbsp;Thus, a patient\u0026rsquo;s history review, including data regarding the family history, blood transfusion, history of anemia, blood loss, melena, and hematochezia, might help conduct a proper investigation for\u0026nbsp;the differential diagnosis\u0026nbsp;of IDA with Thal.\u003c/p\u003e\n\u003cp\u003eIn this study, we utilized only RBC indices and personal demographic data to minimize feature redundancy and enhance the performance of our machine learning model.\u0026nbsp;We demonstrated that MCHC and MCV levels are the two important features for machine learning in the RF and GB models, respectively.\u0026nbsp;MCHC represents the average Hb concentration within a single RBC.\u0026nbsp;Notably, this index is significantly lower in patients with IDA than in patients with Thal\u0026nbsp;(TT and TI) (Table 1).\u0026nbsp;This may be explained by the fact that IDA results from a lack of iron, which is essential for Hb production.\u0026nbsp;As a result, RBCs have lower Hb content and are smaller in size.\u0026nbsp;By contrast, Thal is caused by a defect in globin chain production, with iron supply remaining sufficient.\u0026nbsp;Consequently, MCHC is not as drastically reduced in Thal as in IDA.\u0026nbsp;A recent study used GB to predict individuals with TT, IDA, and a normal\u0026nbsp;condition and\u0026nbsp;identified MCV, MCH, RDW-SD, and Hb levels as the most significant features\u0026nbsp;[34].\u0026nbsp;MCV and MCH are effective markers for screening Thal\u0026nbsp;carriers\u0026nbsp;[35-36].\u0026nbsp;Similarly, in the current study, MCV also emerged as a key feature in the model, thus aligning with previous findings.\u003c/p\u003e\n\u003cp\u003eAdditionally, we compared the diagnostic performance of previously reported formulas with machine learning models\u0026nbsp;(GB and RF)\u0026nbsp;in binary outcomes model.\u0026nbsp;Among the 15 formulas used to predict Thal\u0026nbsp;(TT and TI)\u0026nbsp;and IDA, only the Hct/Hb index demonstrated strong performance.\u0026nbsp;This is the first study to use the Hct/Hb index to discriminate\u0026nbsp;between\u0026nbsp;IDA and Thal\u0026nbsp;(TT and TI).\u0026nbsp;This index is useful for differential diagnosis because patients with Thal can have low Hb levels\u0026nbsp;[37].\u0026nbsp;By contrast, patients\u0026nbsp;with IDA have low RBC production, which contributes to low Hb\u0026nbsp;and Hct levels\u0026nbsp;[38].\u0026nbsp;However, the Hct/Hb index has not been used to differentiate IDA from TT in previous studies.\u0026nbsp;Moreover, the performance of this single index remains lower than\u0026nbsp;that of\u0026nbsp;our machine learning model\u0026nbsp;that uses\u0026nbsp;only RBC indices\u0026nbsp;(Table 4).\u0026nbsp;The remaining 14 formulas demonstrated low performance in our cases because they were originally validated for predicting TT and IDA.\u0026nbsp;However, the Thal group in our study included a diverse range of genotypes that encompasses both TT and TI.\u0026nbsp;Therefore, the applicability of these formulas may be limited in regions where Thal is prevalent.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe demonstrated the efficiency of two predictive models:\u0026nbsp;one distinguishing between\u0026nbsp;Thal and IDA and the\u0026nbsp;other differentiating among Thal, IDA, and the combined group of IDA with Thal by\u0026nbsp;using GB and RF.\u0026nbsp;GB\u0026nbsp;performed better in\u0026nbsp;multiclass outcomes model,\u0026nbsp;with an AUC-ROC of\u0026nbsp;0.910\u0026nbsp;(95%\u0026nbsp;CI:\u0026nbsp;0.859\u0026ndash;0.949).\u0026nbsp;We suggested the\u0026nbsp;application of\u0026nbsp;a predictive model involving three groups for patient care because we could not exclude patients with both IDA and Thal in a real-world clinical setting.\u0026nbsp;However, the accuracy of GB\u0026nbsp;in the testing dataset was lower than that in the training dataset, and this result might have been due to the small sample size.\u0026nbsp;The model performance of GB in the testing dataset remained high and acceptable.\u0026nbsp;However, further prospective studies should be performed to externally validate the performance and refinement of this diagnostic model.\u003c/p\u003e\n\u003cp\u003eFinally, a machine learning approach for discriminating between IDA and Thal using the GB algorithm was developed, along with a web-based prediction tool named\u0026nbsp;\u0026ldquo;PSU Thal-IDA Pred.\u0026rdquo;\u0026nbsp;The probability scores can guide clinicians in selecting suitable confirmation tests in first-visit patients with unknown causes of hypochromic microcytic anemia, resulting in reduced laboratory investigation costs and time and reduced blood volume in the sample collection of patients with anemia.\u0026nbsp;Users can easily access\u0026nbsp;our website at\u0026nbsp;https://srisintornw.shinyapps.io/small_mcv_prediction_v2/.\u0026nbsp;The prediction scores were obtained after\u0026nbsp;inputting the RBC indices.\u003c/p\u003e\n\u003cp\u003eIn conclusion, our study demonstrated that machine learning (GB and RF) algorithms are efficient in discriminating between patients with IDA and Thal but not in complex diseases, such as IDA with Thal. Thus, we recommend the application of this diagnostic model for the diagnosis of IDA and Thal in the Thai population, wherein IDA and Thal are endemic.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was supported funding by the Faculty of Medicine, Prince of Songkla University\u0026nbsp;(REC 62-232-5-2).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eConceptualization:\u0026nbsp;WT, WS, and PS; Methodology, formal analysis, and investigation:\u0026nbsp;WT, WS, TP, and PS; Writing–original draft preparation:\u0026nbsp;WT and WS; Writing–review and editing:\u0026nbsp;WT, WS, TP, and PS; Funding acquisition:\u0026nbsp;WT.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.Some data may not be available because of privacy or ethical reasons.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical approval\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe study was approved by the office of Human Research Ethics Committee, Faculty of Medicine, Prince of Songkla University, and was conducted in accordance with the Declaration of Helsinki (approval number: REC 62-232-5-2). The requirement for inform consent was waived because we used deidentified data.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eNewhall, D. A., Oliver, R. \u0026amp; Lugthart, S. Anaemia: A disease or symptom.\u003cem\u003e Neth\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eJ\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eMed\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e78\u003c/strong\u003e, 104-110 (2020).\u003c/li\u003e\n\u003cli\u003eGBD 2021 Anaemia Collaborators. Prevalence, years lived with disability, and trends in anaemia burden by severity and cause, 1990-2021: findings from the Global Burden of Disease Study 2021. \u003cem\u003eLancet Haematol\u003c/em\u003e. \u003cstrong\u003e10\u003c/strong\u003e, e713-e734 (2023). \u003c/li\u003e\n\u003cli\u003eWinichagoon, P. Prevention and control of anemia: Thailand experiences.\u003cem\u003e J\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eNutr\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e132\u003c/strong\u003e (Supplement), 862S-866S (2002).\u003c/li\u003e\n\u003cli\u003eSirachainan, N. \u003cem\u003eet al\u003c/em\u003e. New mathematical formula for differentiating thalassemia trait and iron deficiency anemia in thalassemia prevalent area: a study in healthy school-age children.\u003cem\u003e Southeast Asian J\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eTrop\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eMed\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003ePublic Health \u003c/em\u003e\u003cstrong\u003e45\u003c/strong\u003e, 174-182 (2014).\u003c/li\u003e\n\u003cli\u003eFucharoen, S. \u0026amp; Winichagoon, P. Haemoglobinopathies in Southeast Asia.\u003cem\u003e Indian J\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eMed\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eRes\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e134\u003c/strong\u003e, 498-506 (2011).\u003c/li\u003e\n\u003cli\u003eJohnson-Wimbley, T. D. \u0026amp; Graham, D. Y. Diagnosis and management of iron deficiency anemia in the 21st century.\u003cem\u003e Ther\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eAdv\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eGastroenterol\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e4\u003c/strong\u003e, 177-184 (2011).\u003c/li\u003e\n\u003cli\u003eNopparatana, C., Nopparatana, C., Saechan, V., Karnchanaopas, S. \u0026amp; Srewaradachpisal, K. Prenatal diagnosis of \u0026alpha;- and \u0026beta;-thalassemias in southern Thailand.\u003cem\u003e Int\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eJ\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eHematol\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e111\u003c/strong\u003e, 284-292 (2020).\u003c/li\u003e\n\u003cli\u003ePornprasert, S., Thongsat, C. \u0026amp; Panyachadporn, U. Evaluation of applying a combination of red cell indexes and formulas to differentiate \u0026beta;-thalassemia trait from iron deficiency anemia in the Thai population.\u003cem\u003e Hemoglobin \u003c/em\u003e\u003cstrong\u003e41\u003c/strong\u003e, 116-119 (2017).\u003c/li\u003e\n\u003cli\u003eAmbayya, A. \u003cem\u003eet al\u003c/em\u003e. Haematological reference intervals in a multiethnic population.\u003cem\u003e PLOS One \u003c/em\u003e\u003cstrong\u003e9\u003c/strong\u003e, e91968 (2014).\u003c/li\u003e\n\u003cli\u003eHuang, T. C. \u003cem\u003eet al\u003c/em\u003e. Discrimination index of microcytic anemia in young soldiers: a single institutional analysis.\u003cem\u003e PLOS One \u003c/em\u003e\u003cstrong\u003e10\u003c/strong\u003e, e0114061 (2015).\u003c/li\u003e\n\u003cli\u003eLaengsri, V. \u003cem\u003eet al\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e ThalPred: a web-based prediction tool for discriminating thalassemia trait and iron deficiency anemia. \u003cem\u003eBMC Med\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eInform\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eDecis\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eMak\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 212 (2019). \u003c/li\u003e\n\u003cli\u003eHasani, M. \u0026amp; Hanani, A. Automated diagnosis of iron deficiency anemia and thalassemia by data mining techniques. \u003cem\u003eIJCSNS \u003c/em\u003e\u003cstrong\u003e17\u003c/strong\u003e, 326-331 (2017).\u003c/li\u003e\n\u003cli\u003e\u0026Ccedil;il, B., Ayyıldız, H. \u0026amp; Tuncer, T. Discrimination of \u0026beta;-thalassemia and iron deficiency anemia through extreme learning machine and regularized extreme learning machine based decision support system. \u003cem\u003eMed\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eHypotheses\u003c/em\u003e \u003cstrong\u003e138\u003c/strong\u003e, 109611 (2020).\u003c/li\u003e\n\u003cli\u003eKarabayir, I., Goldman, S. M., Pappu, S. \u0026amp; Akbilgic, O. Gradient boosting for Parkinson\u0026apos;s disease diagnosis from voice recordings.\u003cem\u003e BMC Med\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eInform\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eDecis\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eMak\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e20\u003c/strong\u003e, 228 (2020).\u003c/li\u003e\n\u003cli\u003eCamaschella, C. Iron-deficiency anemia.\u003cem\u003e N\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eEngl\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eJ\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eMed\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e372\u003c/strong\u003e, 1832-1843 (2015).\u003c/li\u003e\n\u003cli\u003eChong, S. S., Boehm, C. D., Higgs, D. R. \u0026amp; Cutting, G. R. Single-tube multiplex-PCR screen for common deletional determinants of alpha-thalassemia.\u003cem\u003e Blood \u003c/em\u003e\u003cstrong\u003e95\u003c/strong\u003e, 360-362 (2000).\u003c/li\u003e\n\u003cli\u003eJames, G., Witten, D., Hastie, T. \u0026amp; Tibshirani, R. \u003cem\u003eAn Introduction to Statistical Learning\u003c/em\u003e\u003cem\u003e: \u003c/em\u003e\u003cem\u003ewith Applications in R\u003c/em\u003e (Springer US, New York, 2021).https://doi.org/10.32614/CRAN.package.ISLR2.\u003c/li\u003e\n\u003cli\u003eChawla, N. V., Bowyer, K. W., Hall, L. O. \u0026amp; Kegelmeyer, W. P. Smote: synthetic minority over-sampling technique. \u003cem\u003eJ\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eArtifical Intell\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eRes\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e \u003cstrong\u003e16\u003c/strong\u003e, 321-357 (2002).\u003c/li\u003e\n\u003cli\u003ePornprasert, S., Panya, A., Punyamung, M., Yanola, J. \u0026amp; Kongpan, C. Red cell indices and formulas used in differentiation of \u0026beta;-thalassemia trait from iron deficiency in Thai school children.\u003cem\u003e Hemoglobin \u003c/em\u003e\u003cstrong\u003e38\u003c/strong\u003e, 258-261 (2014).\u003c/li\u003e\n\u003cli\u003eNalbantoğlu, B. \u003cem\u003eet al\u003c/em\u003e. Indices used in differentiation of thalassemia trait from iron deficiency anemia in pediatric population: are they reliable?\u003cem\u003e Pediatr\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eHematol\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eOncol\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e29\u003c/strong\u003e, 472-478 (2012).\u003c/li\u003e\n\u003cli\u003eDemir, A., Yarali, N., Fisgin, T., Duru, F. \u0026amp; Kara, A. Most reliable indices in differentiation between thalassemia trait and iron deficiency anemia.\u003cem\u003e Pediatr\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eInt\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e44\u003c/strong\u003e, 612-616 (2002).\u003c/li\u003e\n\u003cli\u003eSirdah, M., Tarazi, I., Al Najjar, E. \u0026amp; Al Haddad, R. Evaluation of the diagnostic reliability of different RBC indices and formulas in the differentiation of the beta-thalassaemia minor from iron deficiency in Palestinian population.\u003cem\u003e Int\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eJ\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eLab\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eHematol\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e30\u003c/strong\u003e, 324-330 (2008).\u003c/li\u003e\n\u003cli\u003eDeLong, ER., DeLong, DM. \u0026amp; Clarke-Pearson, DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. \u003cem\u003eBiometrics\u003c/em\u003e \u003cstrong\u003e44, \u003c/strong\u003e837-845 (1988).\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eR\u003c/em\u003e\u003cem\u003e: \u003c/em\u003e\u003cem\u003eA Language and Environment for Statistical Computing\u003c/em\u003e (R Foundation for Statistical Computing, Vienna, Austria, 2022). https://www.R-project.org.\u003c/li\u003e\n\u003cli\u003eKuhn, M. \u0026amp; Wickham, H. \u003cem\u003eTidymodels\u003c/em\u003e\u003cem\u003e: \u003c/em\u003e\u003cem\u003ea Collection of Packages for Modeling and Machine Learning Using Tidyverse Principles\u003c/em\u003e (2020). https://www.tidymodels.org.\u003c/li\u003e\n\u003cli\u003eWright, M.N. \u0026amp; Ziegler, A. ranger: A Fast Implementation of Random Forests for High Dimensional Data in C++ and R. \u003cem\u003eJ\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eStat\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eSoftw\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e \u003cstrong\u003e77\u003c/strong\u003e, 1-17 (2017).\u003c/li\u003e\n\u003cli\u003eChen, T. \u003cem\u003eet al\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e Xgboost: extreme Gradient Boosting. (2022); 10.32614/CRAN.package.xgboost.\u003c/li\u003e\n\u003cli\u003eThemis, H. E. \u003cem\u003eExtra Recipes Steps for Dealing with Unbalanced\u003c/em\u003e Data. R package. version 1.0.3. (2025). 10.32614/CRAN.package.themis.\u003c/li\u003e\n\u003cli\u003eRobin, X. \u003cem\u003eet al\u003c/em\u003e. pROC: an open-source package for R and S+ to analyze and compare ROC curves. \u003cem\u003eBMC Bioinformatics\u003c/em\u003e. \u003cstrong\u003e12\u003c/strong\u003e, 77 (2011). \u003c/li\u003e\n\u003cli\u003ePansuwan, A., Fucharoen, G., Fucharoen, S., Himakhun, B. \u0026amp; Dangwiboon, S. Anemia, iron deficiency and thalassemia among adolescents in Northeast Thailand: results from two independent surveys.\u003cem\u003e Acta Haematol\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e125\u003c/strong\u003e, 186-192 (2011).\u003c/li\u003e\n\u003cli\u003eNaveed K, \u003cem\u003eet al\u003c/em\u003e\u003cem\u003e. \u003c/em\u003eDefining ferritin clinical decision limits to improve diagnosis and treatment of iron deficiency: A modified Delphi study. \u003cem\u003eInt\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e J\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e Lab\u003c/em\u003e\u003cem\u003e.\u003c/em\u003e\u003cem\u003e Hematol\u003c/em\u003e. \u003cstrong\u003e45\u003c/strong\u003e, 377-386 (2023). \u003c/li\u003e\n\u003cli\u003eCarla, M. G., Rafael, S. P., Isabel, F. G., Cristina, G. F. \u0026amp; Teresa, S. M. New haematologic score to discriminate beta thalassemia trait from iron deficiency anaemia in a Spanish Mediterranean region.\u003cem\u003e Clin\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eChim\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eActa\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cstrong\u003e507\u003c/strong\u003e, 69-74 (2020).\u003c/li\u003e\n\u003cli\u003eZaghloul, A. \u003cem\u003eet al\u003c/em\u003e. Introduction of new formulas and evaluation of the previous red blood cell indices and formulas in the differentiation between beta thalassemia trait and iron deficiency anemia in the Makkah region.\u003cem\u003e Hematology\u003c/em\u003e\u003cem\u003e \u003c/em\u003e\u003cstrong\u003e21\u003c/strong\u003e, 351-358 (2016).\u003c/li\u003e\n\u003cli\u003eWang, W., Ye, R., Tang, B. \u0026amp; Qi, Y. MultiThal-classifier, a machine learning-based multi-class model for thalassemia diagnosis and classification.\u003cem\u003e Clin\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eChim\u003c/em\u003e\u003cem\u003e. \u003c/em\u003e\u003cem\u003eActa \u003c/em\u003e\u003cstrong\u003e567\u003c/strong\u003e, 120025 (2025).\u003c/li\u003e\n\u003cli\u003eYamsri, S. \u003cem\u003eet al\u003c/em\u003e. Prevention of severe thalassemia in northeast Thailand: 16 years of experience at a single university center. \u003cem\u003ePrenat Diagn\u003c/em\u003e. \u003cstrong\u003e30\u003c/strong\u003e,540-546 (2010).\u003c/li\u003e\n\u003cli\u003eChaitraiphop, C.\u003cem\u003e et al\u003c/em\u003e. Thalassemia Screening Using Different Automated Blood Cell Counters: Consideration of Appropriate Cutoff Values. \u003cem\u003eClin Lab\u003c/em\u003e. \u003cstrong\u003e62\u003c/strong\u003e, 545-52 (2016). \u003c/li\u003e\n\u003cli\u003eWeatherall, D. J. \u0026amp; Clegg, J. B. \u003cem\u003eThe Thalassemia Syndromes\u003c/em\u003e. 4th ed (Blackwell Science., Oxford, 2001).\u003c/li\u003e\n\u003cli\u003eLopez, A., Cacoub, P., Macdougall, I. C. \u0026amp; Peyrin-Biroulet, L. Iron deficiency anaemia.\u003cem\u003e Lancet \u003c/em\u003e\u003cstrong\u003e387\u003c/strong\u003e, 907-916 (2016).\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1\u003c/strong\u003e\u003cstrong\u003e.\u003c/strong\u003e Baseline characteristics and hematological features of patients with Thal and IDA.\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"694\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 331px;\"\u003e\n \u003cp\u003eCharacteristic\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eThal\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e(n = 635) (mean \u0026plusmn; SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 100px;\"\u003e\n \u003cp\u003eIDA with Thal (n = 126)\u003c/p\u003e\n \u003cp\u003e(mean \u0026plusmn; SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 95px;\"\u003e\n \u003cp\u003eIDA\u003c/p\u003e\n \u003cp\u003e(n = 382) (mean \u0026plusmn; SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"top\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e-value\u003csup\u003e1\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eSex, female (n, %)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e428 (67.4%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e111 (88.1%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e327 (85.6%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eAge (years)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e52 \u0026plusmn; 21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e41 \u0026plusmn; 15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e47 \u0026plusmn; 19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eRed blood cells (\u0026times;10\u003csup\u003e6\u003c/sup\u003e/\u0026micro;L)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e4.56 \u0026plusmn; 0.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e4.56 \u0026plusmn; 0.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e4.29 \u0026plusmn; 0.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eHemoglobin (g/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e9.64 \u0026plusmn; 1.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e8.67 \u0026plusmn; 2.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e8.65 \u0026plusmn; 1.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eHematocrit (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e30.5 \u0026plusmn; 5.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e28.5 \u0026plusmn; 6.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e29.3 \u0026plusmn; 5.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eMean cell volume (fL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e67.0 \u0026plusmn; 8.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e63.0 \u0026plusmn; 9.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e68.0 \u0026plusmn; 7.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eMean corpuscular hemoglobin (pg)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e21.4 \u0026plusmn; 2.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e19.0 \u0026plusmn; 3.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e20.1 \u0026plusmn; 3.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eMean corpuscular hemoglobin concentration (g/dL)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e31.65 \u0026plusmn; 2.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e30.2 \u0026plusmn; 2.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e29.3 \u0026plusmn; 2.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eRed blood cell distribution width (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e18.6 \u0026plusmn; 4.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 100px;\"\u003e\n \u003cp\u003e20.0 \u0026plusmn; 4.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e19.0 \u0026plusmn; 3.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 70px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"7\" valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eThalassemia type (n)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u0026alpha;-Thal trait\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eHb Constant Spring trait\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eHb H disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eHb H with Constant Spring\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eHb H with Hb E trait\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u0026beta;-Thal trait\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e194\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u0026beta;\u003csup\u003e+\u003c/sup\u003e/\u0026beta;\u003csup\u003e+\u003c/sup\u003e-Thal disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u0026beta;\u003csup\u003e+\u003c/sup\u003e-Thal/Hb E disease\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003eHb E trait\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e218\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e80\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Hb E trait with Hb Constant Spring trait\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; Homozygous Hb E\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e51\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 331px;\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp; HPFH trait\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 94px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 104px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 95px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThal, thalassemia; IDA, iron deficiency anemia; HPFH, hereditary persistence of fetal hemoglobin. \u003csup\u003e1\u003c/sup\u003e: Pearson\u0026rsquo;s chi-squared test; Kruskal\u0026ndash;Wallis rank sum test. \u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e\u003cstrong\u003e.\u003c/strong\u003e Diagnostic performance of the RF and GB models for predicting binary outcomes (Thal [TT and TI] and IDA) in the testing dataset.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"450\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" valign=\"top\" style=\"width: 112px;\"\u003e\n \u003cp\u003eMetric\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 170px;\"\u003e\n \u003cp\u003eGB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"2\" valign=\"bottom\" style=\"width: 168px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 73px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 97px;\"\u003e\n \u003cp\u003e95%\u0026nbsp;CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 72px;\"\u003e\n \u003cp\u003eMedian\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 96px;\"\u003e\n \u003cp\u003e95%\u0026nbsp;CI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSensitivity\u0026nbsp;(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e89.4\u0026ndash;97.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e93.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e89.5\u0026ndash;97.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eSpecificity\u0026nbsp;(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e85.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e78.0\u0026ndash;92.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e85.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e77.6\u0026ndash;93.1\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAccuracy\u0026nbsp;(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e90.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86.8\u0026ndash;94.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e90.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e86.8\u0026ndash;94.6\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eKappa\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.802\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.719\u0026ndash;0.881\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.803\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.717\u0026ndash;0.887\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003eAUC-ROC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.924\u0026ndash;0.982\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\"\u003e\n \u003cp\u003e0.924\u0026ndash;0.982\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eTT,\u0026nbsp;thalassemia trait; TI,\u0026nbsp;Thal\u0026nbsp;intermedia; IDA,\u0026nbsp;iron deficiency anemia; RF,\u0026nbsp;random forest; GB,\u0026nbsp;gradient boosting; AUC-ROC, area under the receiver operating characteristic curve.\u003c/p\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u003c/strong\u003e\u003cstrong\u003e.\u003c/strong\u003e Diagnostic performance of the RF and GB models for predicting multiclass outcomes (Thal [TT and TI], IDA, and IDA with Thal) in the testing dataset.\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"879\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.7397%;\"\u003e\n \u003cp\u003eMetric\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 20.3288%;\"\u003e\n \u003cp\u003eGB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 21.326%;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.7397%;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.9041%;\"\u003e\n \u003cp\u003eThal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.7507%;\"\u003e\n \u003cp\u003eIDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.674%;\"\u003e\n \u003cp\u003eIDA with Thal \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.9041%;\"\u003e\n \u003cp\u003eThal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 7.211%;\"\u003e\n \u003cp\u003eIDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 7.211%;\"\u003e\n \u003cp\u003eIDA with Thal \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.7397%;\"\u003e\n \u003cp\u003eSensitivity,\u0026nbsp;% (95%\u0026nbsp;CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.9041%;\"\u003e\n \u003cp\u003e81.9 (74.8\u0026ndash;88.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.7507%;\"\u003e\n \u003cp\u003e82.1\u0026nbsp;(73.0\u0026ndash;90.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.674%;\"\u003e\n \u003cp\u003e69.1 (50.0\u0026ndash;86.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.9041%;\"\u003e\n \u003cp\u003e89.2 (83.3\u0026ndash;93.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 7.211%;\"\u003e\n \u003cp\u003e76.8 (66.7\u0026ndash;86.8)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 7.211%;\"\u003e\n \u003cp\u003e65.3 (46.4\u0026ndash;84.6)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.7397%;\"\u003e\n \u003cp\u003eSpecificity,\u0026nbsp;% (95%\u0026nbsp;CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.9041%;\"\u003e\n \u003cp\u003e88.5\u0026nbsp;(82.1\u0026ndash;94.3)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.7507%;\"\u003e\n \u003cp\u003e91.7\u0026nbsp;(86.8\u0026ndash;95.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.674%;\"\u003e\n \u003cp\u003e90.3\u0026nbsp;(86.1\u0026ndash;94.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 6.9041%;\"\u003e\n \u003cp\u003e84.5\u0026nbsp;(77.0\u0026ndash;90.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 7.211%;\"\u003e\n \u003cp\u003e93.6\u0026nbsp;(89.6\u0026ndash;96.9)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" style=\"width: 7.211%;\"\u003e\n \u003cp\u003e92.6\u0026nbsp;(88.9\u0026ndash;96.1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.7397%;\"\u003e\n \u003cp\u003eAccuracy,\u0026nbsp;% (95%\u0026nbsp;CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 20.3288%;\"\u003e\n \u003cp\u003e80.4\u0026nbsp;(75.2\u0026ndash;85.2)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 21.326%;\"\u003e\n \u003cp\u003e82.2 (77.0\u0026ndash;87.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.7397%;\"\u003e\n \u003cp\u003eKappa, median\u0026nbsp;(95%\u0026nbsp;CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 20.3288%;\"\u003e\n \u003cp\u003e0.669 (0.586\u0026ndash;0.751)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 21.326%;\"\u003e\n \u003cp\u003e0.689\u0026nbsp;(0.603\u0026ndash;0.770)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" style=\"width: 10.7397%;\"\u003e\n \u003cp\u003eAUC-ROC, median\u0026nbsp;(95%\u0026nbsp;CI)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 20.3288%;\"\u003e\n \u003cp\u003e0.910\u0026nbsp;(0.859\u0026ndash;0.949)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"3\" valign=\"bottom\" style=\"width: 21.326%;\"\u003e\n \u003cp\u003e0.899\u0026nbsp;(0.844\u0026ndash;0.939)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003eThal, thalassemia; TT,\u0026nbsp;Thal\u0026nbsp;trait; TI,\u0026nbsp;Thal\u0026nbsp;intermedia; IDA, iron\u0026nbsp;deficiency anemia; RF,\u0026nbsp;random forest; GB,\u0026nbsp;gradient boosting; AUC-ROC, area under the receiver operating characteristic curve.\u003cstrong\u003e\u003cbr\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4\u003c/strong\u003e\u003cstrong\u003e.\u0026nbsp;\u003c/strong\u003eDiagnostic performance of a single formula and comparison of\u0026nbsp;AUC-ROC\u0026nbsp;between each formula with GB and RF algorithms.\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"960\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 102px;\"\u003e\n \u003cp\u003eFormula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 53px;\"\u003e\n \u003cp\u003eCut-off value\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e95%\u0026nbsp;CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 64px;\"\u003e\n \u003cp\u003eSensitivity\u0026nbsp;(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 60px;\"\u003e\n \u003cp\u003e95%\u0026nbsp;CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003eSpecificity\u0026nbsp;(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e95%\u0026nbsp;CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 57px;\"\u003e\n \u003cp\u003eYouden\u0026rsquo;s index\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e95%\u0026nbsp;CI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eAUC-ROC of single formula\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003eAUC-ROC of GB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003eAUC-ROC of RF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e-value\u003csup\u003e1\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e\u003cem\u003eP\u003c/em\u003e-value\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eMentzer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e16.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e13.0\u0026ndash;19.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e93.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e93.3\u0026ndash;94.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e31.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e27.5\u0026ndash;36.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e25.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e21.5\u0026ndash;29.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.568\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e8.65E-22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e1.55E-23\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eShine \u0026amp; Lal\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e1221.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e942.5\u0026ndash;1499.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e78.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e62.4\u0026ndash;94.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e32.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e14.5\u0026ndash;49.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e10.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e10.0\u0026ndash;11.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.529\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e1.65E-19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e1.57E-21\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eEngland and Fraser\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e9.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e7.3\u0026ndash;10.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e88.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e82.0\u0026ndash;95.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e44.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e34.6\u0026ndash;53.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e32.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e30.4\u0026ndash;35.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.640\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e7.36E-20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e1.01E-18\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eSrivastava\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e5.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e4.6\u0026ndash;5.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e72.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e60.3\u0026ndash;84.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e41.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e34.5\u0026ndash;48.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e14.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e8.8\u0026ndash;19.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.478\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e2.40E-31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e3.16E-30\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eGreen and King\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e82.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e82.5\u0026ndash;82.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e86.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e82.6\u0026ndash;90.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e53.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e49.5\u0026ndash;56.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e39.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e32.0\u0026ndash;47.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.664\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e3.96E-16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e6.33E-15\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eJayabose\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e240.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e232.8\u0026ndash;247.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e92.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e92.5\u0026ndash;92.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e35.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e30.8\u0026ndash;39.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e28.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e23.7\u0026ndash;32.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.596\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e4.27E-20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e5.86E-21\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eRicera\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e3.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e3.1\u0026ndash;3.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e82.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e68.9\u0026ndash;95.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e41.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e25.0\u0026ndash;58.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e23.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e20.0\u0026ndash;27.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.559\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e1.63E-24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e7.07E-24\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eEhsani\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e21.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e14.5\u0026ndash;27.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e70.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e53.3\u0026ndash;88.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e54.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e36.7\u0026ndash;72.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e25.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e25.0\u0026ndash;26.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.580\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e2.44E-21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e7.39E-21\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eSirdah\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e33.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e29.2\u0026ndash;36.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e81.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e65.3\u0026ndash;97.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e48.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e35.1\u0026ndash;62.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e30.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e27.8\u0026ndash;33.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.643\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e4.64E-18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e2.27E-18\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eKerman_i\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e381.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e329.7\u0026ndash;432.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e79.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e65.2\u0026ndash;92.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e35.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e20.3\u0026ndash;51.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e14.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e13.0\u0026ndash;16.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.512\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e7.90E-25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e1.16E-24\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eKerman_ii\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e94.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e75.9\u0026ndash;113.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e73.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e55.2\u0026ndash;92.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e51.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e36.2\u0026ndash;66.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e25.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e21.4\u0026ndash;28.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.579\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e2.45E-19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e1.10E-20\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eKeikhaei\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e24.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e24.4\u0026ndash;24.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e92.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e92.4\u0026ndash;93.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e43.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e41.7\u0026ndash;44.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e36.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e34.1\u0026ndash;37.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.596\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e2.74E-20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e7.02E-21\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eMatos\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e22.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e22.4\u0026ndash;23.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e85.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e82.4\u0026ndash;88.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e50.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e40.2\u0026ndash;59.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e35.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e29.1\u0026ndash;42.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.686\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e1.27E-16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e1.27E-16\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eMCV/Hb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e6.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e6.5\u0026ndash;6.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e89.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e88.9\u0026ndash;90.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e42.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e38.1\u0026ndash;47.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e32.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e27.0\u0026ndash;37.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.635\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e3.85E-20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e4.39E-20\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 102px;\"\u003e\n \u003cp\u003eHct/Hb\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 53px;\"\u003e\n \u003cp\u003e3.2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 77px;\"\u003e\n \u003cp\u003e3.2\u0026ndash;3.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 64px;\"\u003e\n \u003cp\u003e77.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 60px;\"\u003e\n \u003cp\u003e65.0\u0026ndash;90.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e73.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 65px;\"\u003e\n \u003cp\u003e61.5\u0026ndash;86.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 57px;\"\u003e\n \u003cp\u003e51.7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 69px;\"\u003e\n \u003cp\u003e51.0\u0026ndash;52.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.820\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 65px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 63px;\"\u003e\n \u003cp\u003e0.953\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 61px;\"\u003e\n \u003cp\u003e1.12E-07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 69px;\"\u003e\n \u003cp\u003e1.49E-07\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003csup\u003e1\u003c/sup\u003e\u003cem\u003eP\u003c/em\u003e-value of AUC-ROC comparison between single formula and GB, \u003csup\u003e2\u003c/sup\u003e\u003cem\u003eP\u003c/em\u003e-value of AUC-ROC comparison between single formula and RF.\u003c/p\u003e\n\u003cp\u003eGB, gradient boosting; RF, random forest; AUC-ROC, area under the receiver operating characteristic curve.\u003c/p\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"iron deficiency anemia, thalassemia, machine learning, random forest, gradient boosting","lastPublishedDoi":"10.21203/rs.3.rs-5623304/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5623304/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eFormulas based on red blood cell indices have been used to differentiate between iron deficiency anemia (IDA) and thalassemia (Thal). However, such formulas exhibit varying efficiencies. In this study, we aimed to develop a tool for discriminating between IDA and Thal by using the random forest (RF) and gradient boosting (GB) algorithms. Complete blood count data from 1,143 patients with anemia and low mean corpuscular volume were collected (382 patients with IDA, 635 with Thal, and 126 with IDA and Thal). The data were randomly divided into training and testing datasets by using a ratio of 80:20. The RF and GB models had good diagnostic performances for predicting IDA and Thal in the training and testing datasets. In the testing dataset for predicting binary outcomes, GB and RF both had an accuracy of 90.7% and an area under the receiver operating characteristic curve (AUC-ROC) of 0.953. A lower diagnostic performance was observed when patients with IDA and Thal were included. GB and RF showed accuracies of 80.4% and 82.2%, respectively, and AUC-ROC values of 0.910 and 0.899, respectively. A machine learning approach was developed using GB algorithm. This tool may be useful in regions where Thal and IDA are endemic.\u003c/p\u003e","manuscriptTitle":"Machine learning approach for differentiating iron deficiency anemia and thalassemia using random forest and gradient boosting algorithms","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-03-25 10:04:01","doi":"10.21203/rs.3.rs-5623304/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-04-14T06:33:05+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-04-13T12:09:38+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-03-29T11:42:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"263131789791743502969682607727116282760","date":"2025-03-21T06:44:53+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"138834319081823766696498815209441004646","date":"2025-03-21T04:02:25+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-03-20T20:51:42+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-03-20T14:56:35+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2025-03-20T05:57:15+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"eef9be22-942b-469a-8721-36df01cf0c69","owner":[],"postedDate":"March 25th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":46006813,"name":"Health sciences/Diseases/Haematological diseases/Anaemia"},{"id":46006814,"name":"Health sciences/Medical research/Experimental models of disease"}],"tags":[],"updatedAt":"2025-05-19T16:03:07+00:00","versionOfRecord":{"articleIdentity":"rs-5623304","link":"https://doi.org/10.1038/s41598-025-01458-5","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2025-05-15 15:56:59","publishedOnDateReadable":"May 15th, 2025"},"versionCreatedAt":"2025-03-25 10:04:01","video":"","vorDoi":"10.1038/s41598-025-01458-5","vorDoiUrl":"https://doi.org/10.1038/s41598-025-01458-5","workflowStages":[]},"version":"v1","identity":"rs-5623304","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5623304","identity":"rs-5623304","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}