Assessment of machine learning algorithms to predict medical specialty choice

Assessment of machine learning algorithms to predict medical specialty choice | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (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];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Assessment of machine learning algorithms to predict medical specialty choice View ORCID Profile David Vicente Alvarez , View ORCID Profile Milena Abbiati , View ORCID Profile Alban Bornet , View ORCID Profile Georges Savoldelli , View ORCID Profile Nadia Bajwa , View ORCID Profile Douglas Teodoro doi: https://doi.org/10.1101/2025.03.06.25323485 David Vicente Alvarez a University of Geneva Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for David Vicente Alvarez For correspondence: firstname.first_family_name{at}unige.ch firstname.family_name{at}hug.ch Milena Abbiati a University of Geneva b Geneva University Hospitals Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Milena Abbiati Alban Bornet a University of Geneva Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Alban Bornet Georges Savoldelli a University of Geneva b Geneva University Hospitals Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Georges Savoldelli Nadia Bajwa a University of Geneva b Geneva University Hospitals Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Nadia Bajwa For correspondence: firstname.first_family_name{at}unige.ch firstname.family_name{at}hug.ch Douglas Teodoro a University of Geneva Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Douglas Teodoro Abstract Full Text Info/History Metrics Data/Code Preview PDF Abstract Equitable distribution of physicians across specialties is a significant public health challenge. While previous studies primarily relied on classic statistics models to estimate factors affecting medical students’ career choices, this study explores the use of machine learning techniques to predict decisions early in their studies. We evaluated various supervised models, including support vector machines, artificial neural networks, extreme gradient boosting (XGBoost), and CatBoost using data from 399 medical students from medical faculties in Switzerland and France. Ensemble methods outperformed simpler models, with CatBoost achieving a macro AUROC of 76%. Post-hoc interpretability methods revealed key factors influencing predictions, such as motivation to become a surgeon and psychological traits like extraversion. These findings show that machine learning could be used for predicting medical career paths and inform better workforce planning. 1. Introduction The Global Health Workforce Alliance identifies the unequal distribution of medical personnel as one of the foremost public health challenges of the 21st century [ 1 ]. Ensuring an equitable allocation of physicians across various specialties and geographic regions remains a significant challenge for developed nations, many of which rely heavily on internationally trained physicians to meet their healthcare needs [ 2 , 3 ]. Addressing this issue necessitates a coordinated effort to ensure that enough qualified and appropriately specialized physicians are strategically placed to serve regions that are currently underserved. Medical student career choice intentions and motives have a direct effect on physician workforce planning. Individual factors, such as gender, personality, and motivation, have been identified as significantly impacting the dynamics of the career choice process [ 4 ]. Existing studies have focused mostly on identifying factors associated with career choice rather than predicting student career intentions [ 5 – 7 ]; with methods based on classic statistics, such as logistic regression. To the best of our knowledge, only one study applied machine learning methods to predict medical student’s career choices [ 8 ]. However, it focused only on artificial neural networks (ANN). However, ANN tends not to be the best modeling choice for tabular data, being often outperformed by gradient-boosting methods, such as extreme gradient boosting (XGBoost) [ 9 ]. To address this gap, in this study, we aimed to assess several machine learning algorithms for predicting medical students’ career choices. Another contribution of this work is that we investigated post-hoc interpretability methods to identify the factors influencing students’ specialty selection, which could provide hypotheses for factors influencing the career decision process. 2. Methods 2.1. Career choice dataset The career choice dataset contained socio-educational, psychological [ 10 , 11 ], and motivational features for 686 (male=244; female=441) medical students enrolled at the Universities of Geneva (n=285; 41.6%, male=110; female=175), Lausanne (n=141; 20.5%, male=43; female=98), and Strasbourg (n=260; 37.8%, male=91; female=168). Students completed the questionnaire twice: once at the start (year 4) and once at the conclusion of their shared studies (year 6). The outcome labels were defined as the choice of the specialty made by the student in their 6 th year. For simplicity, these labels were separated into 2 by domain experts: “technically oriented specialties” and “person-oriented specialties”: Any missing data point was coded as -1. After data cleaning, including the removal of instances with missing labels or duplicates, the dataset comprised 399 (male=130; female=269) instances. 2.2. Experiment We trained several machine learning classifiers [ 12 ] - logistic regression, decision tree, k-nearest neighbors, support vector machine (SVM), random forest, multi-layer perceptron (MLP), XGBoost, and CatBoost - using a 5-fold cross-validation scheme. In every fold, we used 60% for training, 20% hyper-parameter fine-tuning, and 20% for testing. For hyper-parameter fine-tuning, we did a grid search and used the parameters that led to the best F1-score over the dataset. We then compared model performances against a majority-class classifier, which naively predicted the majority class from the training dataset. We evaluated all models using the testing dataset. We report the average and standard deviation across all folds for the F1-score, area under the receiver operating characteristic curve (AUROC), and accuracy metrics. We used SHAP (SHapley Additive exPlanations) [ 13 ] values to assess the relative importance of different features in predicting medical student career choices. 3. Results and Discussions Table 1 shows the results of our experiments. Among the more sophisticated models, CatBoost and MLP perform comparably, with CatBoost exhibiting the highest AUROC (75.7%) and accuracy (69.2%) while MLP achieved the highest F1-score (67.8%). XGBoost, another gradient-boosting model, performs competitively, with an F1-score of 66.5%, an AUROC of 75.0%, and an accuracy of 68.9%. Despite its simpler computational complexity, the performance of logistic regression is relatively strong, with an F1-score of 64.6% and AUROC of 72.7%, closely followed by the SVM. Random forest has a decent AUROC of 71.4%, however, its overall accuracy and F1-score are lower compared to other models. The kNN and decision tree classifiers are outperformed by most other models. While kNN shows the lowest performance among the non-baseline models, with an F1-score of 52.5% and AUROC of 57.1%, decision tree has moderate success with an F1-score of 56.0% and AUROC of 56.6%. These results indicate that ensemble machine learning algorithms, particularly XGBoost and CatBoost, along with MLPs, outperform simpler models such as decision trees and kNN in predicting the career choices of medical students. The higher performance of ensemble methods is consistent with their well-documented strengths in handling structured tabular data [ 14 ]. View this table: View inline View popup Download powerpoint Table 1. Performance evaluation of all networks 3.1. Explainability Due to the out-of-the-box implementation of interpretability functionalities via SHAP values and the close performance to the best model, CatBoost, we used the XGBoost model to identify features impacting the decision of the model for career prediction. Figure 1 shows the SHAP values for the XGBoost classifier. The left plot summarizes feature importance, while the right plot highlights the impact of features on predicting a student’s choice of “technically oriented specialties.” The Y-axis ranks features by importance, while the X-axis shows SHAP values: positive values push predictions toward choosing a technically oriented specialty, and negative values push them away, with the color gradient indicating feature values (blue = low, pink/red = high). Features such as motivation for becoming a surgeon (“MOT SURGEON”), motivation for general practice (“MOT GENERAL PRACTIONNER”), and the level of the profession of the father (“PROFESSION FATHER”) have the most substantial influence on the model’s output, with higher SHAP values indicating a greater impact. Moreover, high values of motivation for becoming a surgeon push the model’s prediction towards the “technically oriented specialties” class (positive SHAP values), while high values of motivation for becoming a general practitioner decrease the likelihood (negative SHAP values) of choosing a technically oriented specialty. Psychological features like openness (“NEOO”), extraversion (“NEOE”), and self-determined motivation (AMQTOT) play also a relevant role in predicting career decisions. When a high value for the extraversion score is present, the model tends to predict the “technically oriented specialties” class. On the other hand, a high value for openness or self-determined motivation scores pushes the model to predict toward the “person-oriented specialties” class. The educational level of parents (“LEVEL EDU MOTHER” and “LEVEL EDU FATHER”), practice choice (“PRACTICE CHOICE” private vs public domain ), and gender (“SEX”) show limited influence on the model’s decision, which is interestingly divergent from current studies [ 4 ]. Download figure Open in new tab Figure 1. SHAP values with XGBoost classifier. 4. Conclusion In conclusion, this study investigates several machine learning models for predicting medical students’ career choices using data from Swiss and French medical faculties. Gradient-boosting methods achieved the best performance, with accuracy as high as 69%, for predicting between technically and non-technically oriented specialties. Post-hoc interpretability methods highlight the importance of motivation and early specialty preferences in these decisions. Despite these promising results, there is important room for performance improvement, which could be achieved by increasing the training set size and integration of exogenous information. Future work could also focus on investigating the predictive performance of algorithms for more fine-grained career choices. Data Availability Data is not available References 1. ↵ Global strategy on human resources for health: Workforce 2030 [Internet] . [cited 2024 Jun 17 ]. Available from: https://www.who.int/publications/i/item/9789241511131 2. ↵ Bernheim L , Balavoine JF . Pénurie de médecins en Suisse : mythe ou réalité? Rev Med Suisse . 2016 Feb 3; 504 : 280 – 1 . OpenUrl 3. ↵ Ärztezeitung SS . La qualité des soins dépend d’une relève suffisante et qualifiée [Internet] . 2022 [cited 2024 Jun 17 ]. Available from: https://saez.swisshealthweb.ch/fr/article/doi/bms.2022.20632/ 4. ↵ Querido SJ , Vergouw D , Wigersma L , Batenburg RS , De Rond MEJ , Ten Cate OTJ. Dynamics of career choice among students in undergraduate medical courses. A BEME systematic review: BEME Guide No. 33 . Medical Teacher . 2016 Jan 2; 38 ( 1 ): 18 – 29 . OpenUrl CrossRef PubMed 5. ↵ Scott I , Gowans M , Wright B , Brenneis F , Banner S , Boone J. Determinants of choosing a career in family medicine . CMAJ . 2011 Jan 11; 183 ( 1 ): E1 – 8 . OpenUrl Abstract / FREE Full Text 6. Ward AM , Kamien M , Lopez DG . Medical career choice and practice location: early factors predicting course completion, career choice and practice location . Medical Education . 2004 ; 38 ( 3 ): 239 – 48 . OpenUrl PubMed 7. ↵ Singhapricha T , Minkhorst O , Moran T , Swanson J , Shayne P. Predictors of an Initial Academic Position in Emergency Medicine . West J Emerg Med . 2019 Jan ; 20 ( 1 ): 127 – 31 . OpenUrl PubMed 8. ↵ Petek Šter M , Švab I , Šter B. Prediction of intended career choice in family medicine using artificial neural networks . European Journal of General Practice . 2015 Jan 2; 21 ( 1 ): 63 – 9 . OpenUrl PubMed 9. ↵ Chen T , He T , Benesty M , Khotilovich V , Tang Y , Cho H , Chen K , Mitchell R , Cano I , Zhou T , Li M , Xie J , Lin M , Geng Y , Li Y , Yuan J. xgboost: Extreme Gradient Boosting [Internet] . 2014 [cited 2024 Jun 24 ]. p. 1 .7.7.1. Available from: https://CRAN.R-project.org/package=xgboost 10. ↵ Costa Jr PT , McCrae RR . The Revised NEO Personality Inventory (NEO-PI-R) . The SAGE Handbook of Personality Theory and Assessment: Personality Measurement and Testing (Volume 2) . 2008 ; 2 : 179 – 88 . OpenUrl 11. ↵ Spielberger , C.D. , Gorsuch , R.L. , Lushene , R. , Vagg , P.R. , Jacobs , G.A. Manual for the State Trait Anxiety Inventory . Consulting Psychologists Press , Palo Alto . 1983 ; 12. ↵ S Band S , Yarahmadi A , Hsu CC , Biyari M , Sookhak M , Ameri R , Dehzangi I , Chronopoulos AT , Liang HW . Application of explainable artificial intelligence in medical health: A systematic review of interpretability methods . Informatics in Medicine Unlocked . 2023 Jan 1; 40 : 101286 . OpenUrl 13. ↵ Lundberg SM , Lee SI . A Unified Approach to Interpreting Model Predictions . In: Advances in Neural Information Processing Systems [Internet] . Curran Associates, Inc .; 2017 [cited 2024 Jun 17 ]. Available from: https://papers.nips.cc/paper_files/paper/2017/hash/8a20a8621978632d76c43dfd28b67767-Abstract.html 14. ↵ McElfresh D , Khandagale S , Valverde J , Prasad C V , Ramakrishnan G , Goldblum M , White C. When Do Neural Nets Outperform Boosted Trees on Tabular Data? Advances in Neural Information Processing Systems . 2023 Dec 15; 36 : 76336 – 69 . OpenUrl View the discussion thread. Back to top Previous Next Posted March 11, 2025. Download PDF Data/Code Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Assessment of machine learning algorithms to predict medical specialty choice Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Assessment of machine learning algorithms to predict medical specialty choice David Vicente Alvarez , Milena Abbiati , Alban Bornet , Georges Savoldelli , Nadia Bajwa , Douglas Teodoro medRxiv 2025.03.06.25323485; doi: https://doi.org/10.1101/2025.03.06.25323485 Share This Article: Copy Citation Tools Assessment of machine learning algorithms to predict medical specialty choice David Vicente Alvarez , Milena Abbiati , Alban Bornet , Georges Savoldelli , Nadia Bajwa , Douglas Teodoro medRxiv 2025.03.06.25323485; doi: https://doi.org/10.1101/2025.03.06.25323485 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Health Informatics Subject Areas All Articles Addiction Medicine (567) Allergy and Immunology (863) Anesthesia (297) Cardiovascular Medicine (4409) Dentistry and Oral Medicine (443) Dermatology (380) Emergency Medicine (606) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1505) Epidemiology (15205) Forensic Medicine (30) Gastroenterology (1119) Genetic and Genomic Medicine (6573) Geriatric Medicine (666) Health Economics (994) Health Informatics (4511) Health Policy (1365) Health Systems and Quality Improvement (1608) Hematology (537) HIV/AIDS (1263) Infectious Diseases (except HIV/AIDS) (15902) Intensive Care and Critical Care Medicine (1103) Medical Education (620) Medical Ethics (144) Nephrology (665) Neurology (6573) Nursing (345) Nutrition (998) Obstetrics and Gynecology (1139) Occupational and Environmental Health (954) Oncology (3319) Ophthalmology (967) Orthopedics (369) Otolaryngology (420) Pain Medicine (435) Palliative Medicine (129) Pathology (662) Pediatrics (1689) Pharmacology and Therapeutics (691) Primary Care Research (710) Psychiatry and Clinical Psychology (5421) Public and Global Health (9205) Radiology and Imaging (2191) Rehabilitation Medicine and Physical Therapy (1367) Respiratory Medicine (1191) Rheumatology (593) Sexual and Reproductive Health (709) Sports Medicine (529) Surgery (709) Toxicology (99) Transplantation (288) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'9fe7a094f8f2df94',t:'MTc3OTI0MTE2Mw=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();