Predicting the probabilities of missed general practice appointments in England and Wales

Predicting the probabilities of missed general practice appointments in England and Wales | 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 Research Article Predicting the probabilities of missed general practice appointments in England and Wales Morghan Hartmann, Suping Ling, Aimilia Exarchakou, Bernard Rachet, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3836849/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Missing general practice (GP) appointments represent an important challenge for healthcare systems. In England and Wales, reducing the number of missed appointments would benefit both the National Health Service (NHS) and the patients, avoiding delay in diagnosis and treatment. Since the COVID-19 pandemic, appointment mode has shifted substantially, and many GP practices have started scheduling online appointments in place of face-to-face meetings. In this context, our aim was to build and compare prediction models for the probability of missing a GP appointment, as a function of appointment’s characteristics and the level of deprivation of the area where the GP practice is located. We examined all English GP appointments in 2021 and used two different statistical approaches for prediction: a generalized linear model (logistic regression) and a machine learning approach (Extreme gradient boosting). Predictions were further validated with 2022 data. Both approaches provided comparable predictions in term of calibration, with the advantage that results from the logistic regression can be interpreted as odds ratios. Longer time between booking and appointment plays an important role, as well as deprivation. Deprived areas, which already tend to have lower healthcare standards, may also be losing more resources from cancelled and unattended appointments compared to their less deprived counterparts. Investigating the role of contextual factors behind these inequalities (both within and outside the healthcare system) would be an important step forward. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction According to the National Health Service (NHS), 15 million general practice (GP) appointments are wasted each year due to patients failing to alert their practice they will not attend [ 3 ]. This costs NHS approximately £216 million per year. In addition to the financial consequences, patients may also suffer health impacts, such as delay in diagnosis or treatment. A systematic review of patient explanations for missed appointments from 26 studies found that the most common reasons for missing a GP appointment were family commitments, lack of transport, or forgetfulness [ 4 ]. Similarly, the review showed that patients from a low sociodemographic status were most likely to miss appointments. Therefore, any tools that may help GP practices predict which patients are likely to miss their appointments could help GPs in targeting specific individuals for sending reminders, and eventually with re-organisation of their services. This idea is supported by results from a study that found that not receiving an SMS reminder message slightly increased the odds of missing an appointment for adult patients (OR: 1.18; 95%CI: 1.10–1.28) [ 5 ]. Previous work has provided valuable information on factors that influence attendance in medical settings but does not consider the impact of area and GP characteristics. Instead, most of these rely only on patients’ personal characteristics. Many studies describe factors relating to non-attendance but these are often conducted in other countries which do not have the same healthcare system as England or relate to specialty clinics, which would not relate directly to the general practice structure in this country. Having a country-wide analysis for England could help the health system in its entirety predict demand and attendance. In addition, since the COVID-19 pandemic the modes of consultation changed with an acceleration of the use of remote consultation, and we believe those changes are still in place and likely to be the norms at least for the near future. A common approach when aiming to predict missing appointments was to compare both machine learning and traditional methods. Many commonly used algorithms include logistic regression (LR), decision tree (DT), and random forest (RF). In addition to these methods, naïve Bayes (NB), and discriminant analysis (DA), several groups used ensemble methods such as RF and AdaBoost, which are known to rely on fewer independent variables. In this context, our aim was to build and compare prediction models that best describe the probability of missing a GP appointment, as a function of appointment’s characteristics and the level of deprivation of the area where the GP practice is located. Machine Learning Use Tree-based ensemble methods outperformed deep neural networks in no-show prediction in a cardiology clinic which further confirms their usefulness in this particular task [ 6 ]. However, deep learning was useful in predicting attendance in a paediatric primary care clinic when missing data on patient’s characteristics was substantial [ 7 ]. After including local weather information in the prediction model, the accuracy of an out-of-sample test set to predict no-shows was increased from 81–83%, which suggests that this could plays a role in a patient’s decision to attend. Another popular method of prediction was using gradient boosting machines. One study found that a variation of DT, XGBoost, yielded the lowest standard error for predicting appointment demand at two separate hospitals [ 8 ]. Again, XGBoost accurately predicted patient no-shows prospectively (with an Area Under the receiver operating characteristic Curve (AUC) of 0.73), using 39 relevant features and it had the most success compared to other methods (LR, NN, RF, deep learning) [ 9 ]. Since the COVID-19 pandemic, many GP practices have started scheduling online appointments in place of face-to-face meetings. Using a large Chinese hospital’s online appointment records, appointment attendance was predicted using variables such as lead time, day of week, distance, and doctor rating. Machine learning algorithms of LR, k-nearest neighbours (KNN), DT, RF, bagging, and boosting, with bagging achieving the best AUC. In this case, time to appointment and length of patient registration time were most important predictors [ 10 ]. The knowledge from the machine learning models can be used not only for finding variables associated with attendance, but also for scheduling. After finding the most accurate model, Harris et al created an application to optimally schedule appointments for a clinic [ 11 ]. However, in a study predicting missed appointments in a New York-based rural primary care setting, the machine learning classifiers (LR, DT, ensemble learning) did not outperform the multi-stage chain prediction models, which produced 73% accuracy [ 12 ]. This suggests that for some datasets, simple regression-based methods of prediction might produce better results. However, Sotudian et al found that an RF model outperformed other classifiers such as XGBoost, SVM, and LR with AUC of 76%. The most predictive variables of missed breast imaging appointments involved prior appointment history and socioeconomic factors such as income [ 13 ]. Many of these benefitted from having individual patient characteristics, such as age, sex, ethnic background, and socioeconomic status. Another feature of many of these studies is the breakdown of medical specialty. Methods Data Sources We examined all English GP appointments in 2021 [ 14 ] to identify variables that might help predict patient’s missing an appointment. This data, published by NHS Digital, covers participating practices that use EMIS, TPP, Eva Health (Microtest), Informatica, Cegedim (Vision), and Babylon (GP at Hand) information systems. It includes aggregate data on appointment characteristics including appointment date, mode, time between booking and appointment, and healthcare professional type at the level of small administrative areas called Sub-ICB locations (Sub-ICB). This type of data has been publicly available and released monthly since October 2018 with the goal of supporting winter preparedness. We focused on the period starting in January through December 2021, which corresponds to a period after the worst phase of the pandemic. Therefore, we believe that the predictions will not be much affected due to the imposed COVID-19 restrictions. We merged the NHS Digital GP dataset with English indices of deprivation 2019 [ 15 ] to obtain information on the area-level deprivation. This allows for IMD score, quintile, and area population for each Sub-ICB to be included. Validation of the prediction models was completed using the equivalent 2022 GP dataset. Data Pre-processing The variables used were IMD quintile, healthcare professional type seen in the appointment, time between booking and appointment, and appointment mode. Two additional variables were derived to indicate whether the appointment was on a weekend or holiday, which affects attendance based on the time series analysis. Missing values or entries with a ‘data quality’ flag were removed for model training. Analytical methods Generalized Linear Models Generalized Linear Models (GLM), such as logistic regression, are regression models [ 1 ] that can be used for classification tasks. The GLM uses a function to express the expected value as a combination of inputs, which do not need to be linear. One of the main benefits of this method is the ability to interpret regression coefficients which quantify associations between variables (a.k.a features) and the outcome. For the GP dataset, we assumed a binomial distribution for the number of missed appointments among the total number of appointments, since in our specific case, we have aggregate data with counts of appointments, weighted using the number of total appointments. Using frequency weights in a GLM model produces the same results as repeating observations by those frequencies. Training was completed considering four original dataset variables plus two additional variables to encode whether the date was a weekend or holiday. The Python module ‘statsmodels’ was used for fitting the GLM analysis. Extreme Gradient Boosting Extreme gradient boosting (XGBoost) [ 2 ] is a method of gradient boosted trees, available in an open source package. The basis for this method is gradient boosting, which is a machine learning method whose goal is to find a function that approximates the output variable weights based on minimizing a loss function and makes predictions using a series of decision trees. For this model, the standard loss function used is squared error, the number of decision trees was set to 1000, and learning rate was 0.1. This method has seen much success in international machine learning competitions in a wide range of use cases, including sales prediction, event classification, and motion detection. In this case, we use the ‘total appointments’ variable as an input since it will be assigned a weight within the algorithm. Feature importance, specifically gain, is a common metric for XGBoost and other tree-based models. Gradient boosting methods can show which variables are used as tree branches more than others, which indicates importance. This feature importance scoring is referred to as gain. A higher value of gain implies increased importance for making a prediction. Another way of assessing variable effects is by measuring the average training loss reduction gained when using a feature for splitting. Assessing Prediction Performance For accuracy, mean squared error (MSE) and mean absolute percentage error (MAPE) are used. MSE performs the squared difference between observed and predicted value for each individual observation averaged over the total. MAPE evaluate the accuracy as the ratio of the difference in observed and forecasted values (in absolute value) divided by the actual value. This is then summed for every forecasted point in time and averaged. Both metrics help quantifying errors made when using our model-based predictions compared to the observed values (smaller values the better), but the MAPE will be more sensible to errors made when the true value is small as compared to the MSE. To assess the agreement between observed and predicted probabilities of missed appointment, we also looked at the calibration by comparing the predictions obtained using both methods to the observed value for a single non-holiday weekday, varying other pre-specified appointment characteristics. These characteristics were appointment mode, IMD quintile and the number of days between booking and appointment. For investigating the trend according to a given characteristic, we had to fix the others, and the fixed values chosen were face-to-face for the appointment mode, quintile 3 for the IMD, and 8–14 days for the time between booking and appointment. In other words, if we were to compare the prediction with the observed data according to IMD quintiles, we fixed the other variables to face-to-face mode and 8–14 days between booking and appointment, for a single non-holiday weekday. Results Descriptive Statistical Characteristics The analysed dataset contained 284,316,406 appointments (95.3% of the full dataset), while 13,262,722 (4.7%) appointments that were marked ‘unknown’ or ‘data quality’ were excluded. Further information on the data quality flag can be found on the NHS Digital website [ 16 ]. This data is considered experimental, which means that their coverage and quality are not optimised. Many GP practices are not included if they use systems not eligible for inclusion in the dataset. As shown in Table 1 , most appointments in the dataset were attended (95.4%) with 57.2% of all appointments taking place in person. Only 2.5% were booked well in advance (more than 28 days) compared to 45.8% same day. Most patients were seen either by a GP (51.9% of the time) or by another member of practice staff (45.0%). When looking at appointment mode, a greater proportion of missed appointments were categorized as face-to-face (78.1%). More missing appointments tend to be skewed towards lower IMD quintiles compared with attended appointments, which suggests deprivation may play a role. Missed appointments were also made normally days in advance, while the majority of attended appointments happened on the same day they were booked (47%). By plotting the percentage of daily missed appointments in 2021, we saw clear patterns in face-to-face, telephone, and video conference/online modes (Fig. 1 ). Each peak occurs on the weekend, due to the decreased number of appointments available during these days. In video conference mode, the number of peaks in missed appointments increases in the final quarter of 2021. Less of a pattern is seen in the home visit category, with random peaks occurring over the course of the year. Autumn and winter months see an increase in the percentage of weekend missed appointments, which is not related to the total number of appointments, since there is actually an increase in appointments made in winter months. We examined trends in missed GP appointments for different sub-ICBs. Figure 2 shows the contrast between area-level deprivation and the proportion of missed appointments by area in England. Places with high deprivation in the Northwest seem to have high rates of missing appointments. However, low deprivation areas in the Midlands and Southeast seem to also have high missing rates. Therefore, visually, we have no clear relationship between IMD score and appointment attendance. Clusters of higher missed percentages were found in areas containing large cities, such as London, Birmingham, and Manchester. These are cities that tend to be more deprived and have unequal access to medical resources. When stratifying the data based on appointment mode we find that face-to-face appointments in all areas have a higher likelihood of being missed compared to other modes. Video conference calls also have a higher percentage of missed appointments in most areas compared to home visits and telephone calls. Multivariable Analyses: Odds Ratios and Feature Importance Table 2 shows the associated odds ratios of missed appointment for each of the variables. One can see a clear pattern where an increase in number of days between booking and appointment corresponds to an increase in the probability of missing the appointment. IMD status seems to influence appointment missingness, since there is decreased odds as deprivation decreases (OR 0.7). It is interesting to note that our model found that non-GP health professionals have an increased odds of having their appointment missed. If days of the week are considered individually, we find that Monday through Friday have similar percent missed rates. This is why we have considered only whether the day was a holiday or weekend. Holidays and weekends are highly associated with increased rate of missed appointments, which can be seen in the initial time series analysis as well. The next step to look at trends of missingness across different sub-ICB locations to see if certain areas are more prone to high missing rates due to medical access. For XGBoost, when measuring loss reduction gain, we find that total appointments, appointment mode, very short and very long time between booking and appointment, and whether or not it was a weekend had the most success (Fig. 3 ). Time Series Prediction Figure 4 shows the results of using the GLM and XGBoost models fitted to 2021 data to predict the summer seasonal results of 2022 data. This graph was obtained by averaging all the predicted values over the course of a day into a single daily value. As seen, GLM performs well with the majority of days, but underestimates weekday holiday cases. Since XGBoost does not have linear fitting, it performs better than GLM when accounting for holidays and weekends. However, due to the nonlinear fitting, the danger of overfitting increases. The only date that proves difficult for XGBoost to predict was a bank holiday that was not in the 2021 training dataset (19 Sept). Therefore, if using this as a prediction tool, it would be important to carefully label any holidays that did not occur in previous years. Once we make predictions based on the test dataset, which comes from a different year (2022), we obtain both MSE or MAPE for both methods (Table 3 ). MSE and MAPE should ideally be small and close to zero. In terms of individual predictions based on the training features, XGBoost outperforms GLM for the percentage missed appointments over the year. However, this is due in part to the high percent error for the holiday dates that GLM is unable to predict. When training and testing on weekdays only, the overall MSE and MAPE decreases for both methods. Calibration In Fig. 5 , we observe the results of predicting percent missed appointments on a single weekday, 1st of March 2022. GLM and XGBoost have similar accuracy results when comparing the predictions to the actual data, but GLM obtains higher accuracy in two out of the three categories than the XGBoost model. However, days are highly variable, so it is difficult to come to a conclusion overall on which method performs better. For example, when predicting results for July 1, 2021, XGBoost outperforms GLM. Generally, GLM seems to better estimate the effect of varying appointment mode, while XGBoost has higher accuracy when predicting the effect of time between booking and appointment. Further tuning of both models is needed for predicting home visit appointment attendance, since both greatly overestimated the percent missed. From the results from this single day, the predicted probability of a patient missing an appointment (reference values: face-to-face appointment, IMD-3 and 8–14 days between booking and appointments) seems to correlate directly with the time between booking and appointment and increased deprivation. Without stratifying by appointment mode and other factors, it is impossible to know whether this is due to factors relating to IMD status. Both Face-to-face and video conference appointment modes have similar attendance rates, while telephone mode has lower predicted rates. Using this information, GP practices could send reminders in the immediate lead-up to an appointment made more than a week before for a better attendance rate. We also compare the performance of single day prediction of a model trained on all of the data versus a model trained only on weekday data. When predicting this single day using the weekday only model, the results yield slightly higher accuracies, but increased testing on other daily data would be needed to confirm this. Discussion Previous work has explored the factors in missing appointments and suggested potential ways to improve the rate of patient involvement. Many of these studies benefit from having individual-level information about each appointment that was not available for our research. However, our work uses a dataset that contains information for the whole of England, which provides insight into trends on a larger scale and on actionable health system components. We found that missed GP appointments can be predicted by appointment characteristics, such as mode, time between booking and appointment, type of day (holiday or weekend). Higher odds of missing an appointment were due to a long delay between the appointment being booked and the appointment actually taking place, and the fact that the consultation was planned with a non-GP health professional. We also estimated that deprivation affects the probability of missing an appointment, with lower probabilities of missing appointment for the least deprived areas. However, these probabilities are still non negligeable in least deprived areas, which reflects that there are other shared factors between areas (more or least deprived) that are not accounted for in this work, such as accessibility of consultations, opening hours of the GP practice, etc. The potential role of accessibility seems also supported by the lower odds of missing appointment with telephone consultation mode. In a study conducted in West Yorkshire, adult patients who missed a GP appointment and their counterparts who were present were sent a questionnaire asking about factors contributing to their decision to attend or not attend [ 17 ]. Over 40% of the patients who missed appointments said they had forgotten, but others cited personal commitments or illness. This seems to agree with the results we found, as a strong predictor of missing an appointment was time between booking and appointment. Within one NHS healthcare trust, the highest predictor of unkept appointments was whether a patient had already missed one within the previous year [ 18 ]. Unfortunately, this information was not available for our particular work, but could be useful here. For a cohort analysis in 11 Scottish NHS health boards, patient data was used to analyse differences between patients who miss multiple appointments versus a few [ 19 ]. It was found that older patients (76–90 years) were most likely to have a high proportion of missed appointments. More urban practices had a higher risk of missed appointments compared to rural areas which was also shown in our study. Again, these results suggest that including more details about the individual making the appointment and their interaction with the GP characteristics, would contribute to prediction success. Therefore, one limitation of our work is the lack of information on patient and practice characteristics. Further investigation is needed into the mechanism behind the increased odds of missing appointment when a non-GP health professional was being seen. The results from this and previous work suggest that practice-level modifications might be useful in preventing time and resource loss. For example, since most missed appointments are scheduled 2–3 days earlier, these could be flagged as high risk and multiple reminders could be sent by text on the mobile phone. Further research is needed in meaningful ways of engaging patients who miss their scheduled appointments and ways to encourage local areas to collect appointment statistics. Another important consideration relates to deprived populations. A diabetes clinic found that their non-attending patients tended to be single parents, which supports the hypothesis that individual social factors associated with deprivation might contribute to non-attendance probability [ 20 ]. Similarly, an Exeter-based GP found that low socioeconomic status and younger age were predictors of non-attendance. These findings might suggest that transport difficulties or job responsibilities might be reasons for not attending [ 21 ]. One machine learning-based study found that outpatient appointments in Wales with a higher Townsend Index (more deprived) tended to have a higher “did not attend” rate [ 22 ]. Our results seem to agree with these findings as IMD status as the odds of missing an appointment increases as deprivation increases. Therefore, an implication of this work is that impoverished areas, which already tend to have lower healthcare standards, may also be losing more resources from cancelled and unattended appointments than their less deprived counterparts. However, we have not been able to investigate the role of other contextual factors (both within and outside the healthcare system). Such enhanced model will help reduce missing appointments and improve outcomes for patients. This will require future collaboration with hospitals, health care professionals, and patients [ 23 ]. Conclusion This work showed the utility of using GLM and XGBoost to predict patient attendance in GP practices across England sub-ICB locations. This work has many strengths, including the extent of areas and deprivation levels across England that are featured in the analysis. The ability of XGBoost to predict time series trends in missing appointments was a key finding, as well as good calibration for both approaches. No approach clearly outperforms, and more work is needed in refining the model by adding additional variables or by comparing its performance to other ML methods. Newer 2022 practice-level data available on NHS digital website includes reason for attendance, which could improve the prediction model’s ability. However, this attendance data is only available at practice level. In addition, other sources of information, such as population, hospital or trust characteristics could be merged to see the influence of resources on GP attendance. Overall, the results of this work imply a large amount of NHS resource every year is lost across the whole of England due to individuals missing appointments. This is even more important for deprived areas, but it is crucial to consider that predictors in missed appointments should not be used to further prevent deprived groups from obtaining GP appointments. Instead, we should strive to improve health access for everyone by applying interventions that might decrease non-attendance. With further research into healthcare system prediction, there is an opportunity to better prepare for non-attendance in GP practices which may improve care quality and efficiency. Declarations CRediT authorship contribution statement MH: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Software, Validation, Visualization, Writing – original draft, Writing – review & editing. SL, AE and BR : Conceptualization, Investigation, Writing – review & editing. AB: Conceptualization, Investigation, Methodology, Supervision, Project administration, Validation, Writing – review & editing. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Funding information Inequalities in Cancer Outcome Network is funded by Cancer Research UK programme (Grant No. EPNCZS34). Ethics approval and consent to participate Not applicable. Data availability Data are freely publicly available on the NHS website https://digital.nhs.uk/data-and-information/publications/statistical/appointments-in-general-practice References A. J. Dobson and A. G. Barnett, An Introduction to Generalized Linear Models, 2018. T. Chen and C. Guestrin, "XGBoost: A Scalable Tree Boosting System," in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016. NHS, "Missed GP appointments costing NHS millions," 2019. J. Parsons, C. Bryce and H. Atherton, "Which patients miss appointments with general practice and the reasons why: a systematic review," British Journal of General Practice, vol. 71, no. 707, pp. e406-e412, 2021. J. Lorenz and K. Hawkins, "Getting to Know the No-Show: Predictive Modeling of Missing a Medical Appointment," SAS Proceedings, 2018. S. Srinivas and H. Salah, "Consultation length and no-show prediction for improving appointment scheduling efficiency at a cardiology clinic: A data analytics approach," International Journal of Medical Informatics, vol. 145, 2021. D. Liu, W.-Y. Shin, E. Sprecher, K. Conroy, O. Santiago, G. Wachtel and M. Santillana, "Machine learning approaches to predicting no-shows in pediatric medical appointment," NPJ Digital Medicine, vol. 5, 2022. B. Klute, A. Homb, W. Chen and A. Stelpflug, "Predicting Outpatient Appointment Demand Using Machine Learning and Traditional Methods," Journal of Medical Systems, 2019. S. Rothenberg, B. Bame and E. Herskovitz, "Prospective Evaluation of a Machine-Learning Prediction Model for Missed Radiology Appointments," Journal of Digital Imaging, vol. 35, pp. 1690-1693, 2022. G. Fan, Z. Deng, Q. Ye and B. Wang, "Machine learning-based prediction models for patients no-show in online outpatient appointments," Data Science and Management, vol. 2, pp. 45-52, 2021. S. L. Harris and M. Samorani, "On selecting a probabilistic classifier for appointment no-show prediction," Decision Support Systems, vol. 142, 2021. L. A. Lekham, Y. Wang, E. Hey, S. S. Lam and M. T. Khasawneh, "A Multi-Stage predictive model for missed appointments at outpatient primary care settings serving rural areas," IISE Transactions on Healthcare Systems Engineering, vol. 11, no. 2, 2021. S. Sotudian, A. Afran, C. A. LeBedis, A. F. Rives, I. C. Paschalidis and M. D. Fishman, "Social determinants of health and the prediction of missed breast imaging appointments," BMC Health Services Research, vol. 22, 2022. NHS Digital, "Appointments in General Practice," 24 November 2022. [Online]. Available: https://digital.nhs.uk/data-and-information/publications/statistical/appointments-in-general-practice. GOV.UK, "English indices of deprivation 2019," 26 September 2019. [Online]. Available: https://www.gov.uk/government/statistics/english-indices-of-deprivation-2019. [Accessed 2022]. N. Digital, "Appointments in general practice: supporting information," 2023 July 26. [Online]. Available: https://digital.nhs.uk/data-and-information/publications/statistical/appointments-in-general-practice/appointments-in-general-practice-supporting-information#data-quality. R. D. Neal, M. Hussain-Gambles, V. L. Allgar, D. A. Lawlor and O. Dempsey, "Reasons for and consequences of missed appointments in general practice in the UK: questionnaire survey and prospective review of medical records," BMC Family Practice, 2005. S. Philpott-Morgan, D. B. Thakrar, J. Symons, D. Ray, H. Ashrafian and A. Darzi, "Characterising the nationwide burden and predictors of unkept outpatient appointments in the National Health Service in England: A cohort study using a machine learning approach," PLOS Medicine, 2021. D. A. Ellis, R. McQueenie, A. McConnachie, P. Wilson and A. Williamson, "Demographic and practice factors predicting repeated non-attendance in primary care: a national retrospective cohort analysis," The Lancet Public Health, vol. 2, no. 12, pp. e551-e559, 2017. P. H. Dyer, C. E. Lloyd, R. J. Lancashire, S. C. Bain and A. H. Barnett, "Factors associated with clinic non-attendance in adults with type 1 diabetes mellitus," Diabetic Medicine, vol. 15, no. 4, pp. 339-43, 1998. W. Hamilton, A. Round and D. Sharp, "Patient, hospital, and general practitioner characteristics associated with non-attendance: a cohort study," British Journal of General Practice, vol. 52, no. 477, 2002. E. Incze, P. Holborn, G. Higgs and A. Ware, "Using machine learning tools to investigate factors associated with trends in ‘no-shows’ in outpatient appointments," Health & Place, vol. 67, 2021. F. Grimm, "Predicting missed hospital appointments using machine learning - what are the risks?," Medium, 2 August 2019. [Online]. Tables Table 1 Descriptive Statistics for GP Appointments in England and Wales, 2021. N (%) Total: 297,579,128 Missed Appointments (%) Total: 12,964,247 Attended Appointments (%) Total: 284,614,881 Healthcare Professional Type GP 154,300,596 (51.9) 3,790,637 (29.2) 150,509,959 (52.9) Other Practice Staff 133,973,430 (45.0) 8,876,611 (68.5) 125,096,819 (44.0) Unknown 9,305,102(3.1) 296,999 (2.3) 9,008,103 (3.2) Appointment Mode Face-to-Face 170,511,754 (57.2) 10,128,862 (78.1) 160,382,892 (56.4) Home Visit 1,494,101 (0.5) 81,448 (0.62) 1,412,653 (0.5) Telephone 114,574,823 (38.5) 2,274,779 (17.5) 112,300,044 (39.5) Video 1,479,434 (0.5) 69,528 (0.54) 1,409,906 (0.5) Unknown 9,519,016 (3.2) 409,630 (3.2) 9,109,386 (3.2) Time Between Booking and Appointment Same Day 136,224,901 (45.8) 2,376,354 (18.3) 133,848,547 (47.0) 1 Day 27,784,950 (9.3) 1,121,970 (8.7) 26,662,980 (9.4) 2 to 7 Days 64,510,478 (21.7) 3,852,373 (29.7) 60,658,105 (21.3) 8 to 14 Days 35,680,569 (12.0) 2,667,503 (20.6) 33,013,066 (11.6) 15 to 21 Days 16,728,556 (5.6) 1,359,819 (10.5) 15,368,737 (5.4) 22 to 28 Days 9,228,168 (3.1) 793,801 (6.1) 8,434,367 (3.0) More than 28 Days 7,301,308 (2.5) 786,870 (6.1) 6,514,438 (2.3) Unknown 120,198 (0.04) 5,557 (0.04) 114,641 (0.04) IMD 2019 Quintile 1 – most deprived 56,666,442 (19.0) 2,965,739 (22.9) 53,700,703 (18.9) 2 50,226,054 (16.9) 2,207,680 (17.0) 48,018,374 (16.9) 3 80,184,037 (26.9) 3,405,738 (26.3) 76,778,299 (27.0) 4 61,785,221 (20.8) 2,532,201 (19.5) 59,253,020 (20.8) 5 – least deprived 48,717,374 (16.4) 1,852,889 (14.3) 46,864,485 (16.5) Table 2 Adjusted Odds Ratio of missed GP appointment (England & Wales, 2021) estimated with the Generalised Linear Model (all variables detailed in the table were included as predictors in the model) Odds Ratio (OR) All Days Weekdays Only Healthcare Professional Type GP (reference category) 1 1 Other Practice Staff 1.669 (1.666–1.672) 1.754 (1.751–1.756) Appointment Mode Face-to-Face (reference category) 1 1 Home Visit 0.910 (0.901–0.918) 0.938 (0.929–0.946) Telephone 0.502 (0.501–0.503) 0.528 (0.527–0.529) Video 0.913 (0.907–0.920) 0.928 (0.920–0.936) Time Between Book and Appointment Same Day 0.497 (0.496–0.499) 0.516 (0.515–0.517) 1 Day (reference category) 1 1 2 to 7 Days 1.390 (1.388–1.394) 1.372 (1.369–1.374) 8 to 14 Days 1.675 (1.672–1.678) 1.659 (1.655–1.664) 15 to 21 Days 1.825 (1.820–1.829) 1.804 (1.799–1.810) 22 to 28 Days 1.890 (1.885–1.896) 1.873 (1.868–1.879) More than 28 Days 2.227 (2.221–2.234) 2.238 (2.232–2.246) IMD 2019 Quintile 1 – most deprived (reference category) 1 1 2 0.835 (0.834–0.836) 0.842 (0.841–0.844) 3 0.782 (0.781–0.783) 0.763 (0.761–0.764) 4 0.789 (0.787–0.791) 0.767 (0.765–0.768) 5 – least deprived 0.700 (0.699–0.702) 0.673 (0.672–0.674) Date Characteristics Weekday (reference category) 1 1 Holiday 0.973 (0.966–0.977) 1.00 (0.998–1.01) Weekend 2.918 (2.895–2.942) -- Table 3 Mean Squared Error and Mean Absolute Percentage Error for both prediction methods, England & Wales, 2021 All Data Weekdays Only MSE (%) MAPE (%) MSE MAPE(%) GLM 44.5 51.1 29.3 44.9 XGBoost 26.8 48.1 16.0 40.7 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted 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-3836849","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":266039107,"identity":"5ee6fba3-96fb-40de-b6c5-c2cf3b7b9716","order_by":0,"name":"Morghan Hartmann","email":"","orcid":"","institution":"London School of Hygiene \u0026 Tropical Medicine","correspondingAuthor":false,"prefix":"","firstName":"Morghan","middleName":"","lastName":"Hartmann","suffix":""},{"id":266039108,"identity":"30062d88-4adc-46f9-9c95-1c44bd66d360","order_by":1,"name":"Suping Ling","email":"","orcid":"","institution":"London School of Hygiene \u0026 Tropical Medicine","correspondingAuthor":false,"prefix":"","firstName":"Suping","middleName":"","lastName":"Ling","suffix":""},{"id":266039109,"identity":"04051906-a856-4867-94c1-8a409e102253","order_by":2,"name":"Aimilia Exarchakou","email":"","orcid":"","institution":"London School of Hygiene \u0026 Tropical Medicine","correspondingAuthor":false,"prefix":"","firstName":"Aimilia","middleName":"","lastName":"Exarchakou","suffix":""},{"id":266039110,"identity":"c8bcb42e-a8da-47d1-ab4a-74abefd68e73","order_by":3,"name":"Bernard Rachet","email":"","orcid":"","institution":"London School of Hygiene \u0026 Tropical Medicine","correspondingAuthor":false,"prefix":"","firstName":"Bernard","middleName":"","lastName":"Rachet","suffix":""},{"id":266039111,"identity":"aaf42428-cea8-4194-95a7-64850f4ef298","order_by":4,"name":"Aurélien Belot","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABD0lEQVRIie3Rv0vDQBTA8RcKcTl0PWlI/oUnglPI3/KOg3PxF7h2KBQ62vXAf8I/4UowLtGuBzqkS+dzkQ6iXvAHCEnr6HDfIeSGD+8lBxAK/dca9A8OkQHIE//yFdtA6JOAAVIM+OAvBH5IuZ3s6dNlQxf5GQwnxjyvFyy7nt3ewKgArE0n4fb4EAnVJSQVzTU9MnwqYwuVBLwfd4+xKuaEpRjzEyxZS7j0JDaAi26RWbWzJnz35Nx58sAy3ZK3foJ+iv9jpp0CnhgG1pNoanoXO6hXA7+YFNOkwrlWkqGVR1ZcSbbf8/npnYqcey3EbDhZNi4v0kyLlXUvRbpbU/dm38W/TrT5IkOhUCi0pQ+lDWBRAuPx6wAAAABJRU5ErkJggg==","orcid":"","institution":"London School of Hygiene \u0026 Tropical Medicine","correspondingAuthor":true,"prefix":"","firstName":"Aurélien","middleName":"","lastName":"Belot","suffix":""}],"badges":[],"createdAt":"2024-01-05 09:29:43","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3836849/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3836849/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":49440584,"identity":"e90c004f-a259-4de2-a260-e260bc100dbd","added_by":"auto","created_at":"2024-01-10 21:54:51","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":211938,"visible":true,"origin":"","legend":"\u003cp\u003ePercent missed GP appointments by appointment mode in 2021 weekdays averaged over all Sub-Integrated Care Board Locations in England \u0026amp; Wales\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-3836849/v1/93569268307d85de43dc65c0.png"},{"id":49440586,"identity":"1948af44-5e58-4c46-9aca-6c110c0d9590","added_by":"auto","created_at":"2024-01-10 21:54:51","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":315514,"visible":true,"origin":"","legend":"\u003cp\u003eIMD Score lowest (least deprived) to highest (most deprived) (left) and Percent missed GP appointments in Sub-Integrated Care Board Locations in England \u0026amp; Wales (right)\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-3836849/v1/f500e9c0a2fc1a26c92d3a87.png"},{"id":49440585,"identity":"5ae8a692-8eb6-4681-8efd-77867d824451","added_by":"auto","created_at":"2024-01-10 21:54:51","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":22460,"visible":true,"origin":"","legend":"\u003cp\u003eVariable importance scores in XGBoost model for model trained on all data and weekdays only (England \u0026amp; Wales, 2021)\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-3836849/v1/de8b0abfea326ac8723cb871.png"},{"id":49440588,"identity":"2cd36ca4-dfd6-4329-8b9d-3510d775ad40","added_by":"auto","created_at":"2024-01-10 21:54:52","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":315727,"visible":true,"origin":"","legend":"\u003cp\u003eTime series prediction of missed GP appointments trained on all days (top) and only weekdays (bottom) and averaged over area and appointment characteristics, England \u0026amp; Wales, 2021\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-3836849/v1/8dec33710087ff5b46c09f79.png"},{"id":49440587,"identity":"0cdd1726-4f7d-4f09-910e-3ad5c4bd8d81","added_by":"auto","created_at":"2024-01-10 21:54:51","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":396257,"visible":true,"origin":"","legend":"\u003cp\u003ePercent missing predictions for a singular non-holiday, weekday in March 2022 while varying a) IMD status, b) appointment mode, and c) time between booking and appointment trained on all days (first column) and only weekdays (second column). While varying one variable, other appointment characteristics were fixed as face-to-face, IMD 3, and 8-14 days between booking and appointment, England and Wales.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-3836849/v1/781c9c46e765628dc95e87ef.png"},{"id":49577216,"identity":"68426ec8-4648-4a6c-b9cf-77f973d0e0b0","added_by":"auto","created_at":"2024-01-14 03:37:20","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1153958,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3836849/v1/9082186f-9397-4c52-9cd3-dcf5e9013976.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Predicting the probabilities of missed general practice appointments in England and Wales","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAccording to the National Health Service (NHS), 15\u0026nbsp;million general practice (GP) appointments are wasted each year due to patients failing to alert their practice they will not attend [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. This costs NHS approximately \u0026pound;216\u0026nbsp;million per year. In addition to the financial consequences, patients may also suffer health impacts, such as delay in diagnosis or treatment.\u003c/p\u003e \u003cp\u003eA systematic review of patient explanations for missed appointments from 26 studies found that the most common reasons for missing a GP appointment were family commitments, lack of transport, or forgetfulness [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Similarly, the review showed that patients from a low sociodemographic status were most likely to miss appointments. Therefore, any tools that may help GP practices predict which patients are likely to miss their appointments could help GPs in targeting specific individuals for sending reminders, and eventually with re-organisation of their services. This idea is supported by results from a study that found that not receiving an SMS reminder message slightly increased the odds of missing an appointment for adult patients (OR: 1.18; 95%CI: 1.10\u0026ndash;1.28) [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003ePrevious work has provided valuable information on factors that influence attendance in medical settings but does not consider the impact of area and GP characteristics. Instead, most of these rely only on patients\u0026rsquo; personal characteristics. Many studies describe factors relating to non-attendance but these are often conducted in other countries which do not have the same healthcare system as England or relate to specialty clinics, which would not relate directly to the general practice structure in this country. Having a country-wide analysis for England could help the health system in its entirety predict demand and attendance. In addition, since the COVID-19 pandemic the modes of consultation changed with an acceleration of the use of remote consultation, and we believe those changes are still in place and likely to be the norms at least for the near future.\u003c/p\u003e \u003cp\u003eA common approach when aiming to predict missing appointments was to compare both machine learning and traditional methods. Many commonly used algorithms include logistic regression (LR), decision tree (DT), and random forest (RF). In addition to these methods, na\u0026iuml;ve Bayes (NB), and discriminant analysis (DA), several groups used ensemble methods such as RF and AdaBoost, which are known to rely on fewer independent variables.\u003c/p\u003e \u003cp\u003eIn this context, our aim was to build and compare prediction models that best describe the probability of missing a GP appointment, as a function of appointment\u0026rsquo;s characteristics and the level of deprivation of the area where the GP practice is located.\u003c/p\u003e\n\u003ch3\u003eMachine Learning Use\u003c/h3\u003e\n\u003cp\u003eTree-based ensemble methods outperformed deep neural networks in no-show prediction in a cardiology clinic which further confirms their usefulness in this particular task [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. However, deep learning was useful in predicting attendance in a paediatric primary care clinic when missing data on patient\u0026rsquo;s characteristics was substantial [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. After including local weather information in the prediction model, the accuracy of an out-of-sample test set to predict no-shows was increased from 81\u0026ndash;83%, which suggests that this could plays a role in a patient\u0026rsquo;s decision to attend. Another popular method of prediction was using gradient boosting machines. One study found that a variation of DT, XGBoost, yielded the lowest standard error for predicting appointment demand at two separate hospitals [\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Again, XGBoost accurately predicted patient no-shows prospectively (with an Area Under the receiver operating characteristic Curve (AUC) of 0.73), using 39 relevant features and it had the most success compared to other methods (LR, NN, RF, deep learning) [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSince the COVID-19 pandemic, many GP practices have started scheduling online appointments in place of face-to-face meetings. Using a large Chinese hospital\u0026rsquo;s online appointment records, appointment attendance was predicted using variables such as lead time, day of week, distance, and doctor rating. Machine learning algorithms of LR, k-nearest neighbours (KNN), DT, RF, bagging, and boosting, with bagging achieving the best AUC. In this case, time to appointment and length of patient registration time were most important predictors [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]. The knowledge from the machine learning models can be used not only for finding variables associated with attendance, but also for scheduling. After finding the most accurate model, Harris et al created an application to optimally schedule appointments for a clinic [\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. However, in a study predicting missed appointments in a New York-based rural primary care setting, the machine learning classifiers (LR, DT, ensemble learning) did not outperform the multi-stage chain prediction models, which produced 73% accuracy [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]. This suggests that for some datasets, simple regression-based methods of prediction might produce better results. However, Sotudian et al found that an RF model outperformed other classifiers such as XGBoost, SVM, and LR with AUC of 76%. The most predictive variables of missed breast imaging appointments involved prior appointment history and socioeconomic factors such as income [\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eMany of these benefitted from having individual patient characteristics, such as age, sex, ethnic background, and socioeconomic status. Another feature of many of these studies is the breakdown of medical specialty.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eData Sources\u003c/h2\u003e \u003cp\u003eWe examined all English GP appointments in 2021 [\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] to identify variables that might help predict patient\u0026rsquo;s missing an appointment. This data, published by NHS Digital, covers participating practices that use EMIS, TPP, Eva Health (Microtest), Informatica, Cegedim (Vision), and Babylon (GP at Hand) information systems. It includes aggregate data on appointment characteristics including appointment date, mode, time between booking and appointment, and healthcare professional type at the level of small administrative areas called Sub-ICB locations (Sub-ICB). This type of data has been publicly available and released monthly since October 2018 with the goal of supporting winter preparedness.\u003c/p\u003e \u003cp\u003eWe focused on the period starting in January through December 2021, which corresponds to a period after the worst phase of the pandemic. Therefore, we believe that the predictions will not be much affected due to the imposed COVID-19 restrictions. We merged the NHS Digital GP dataset with English indices of deprivation 2019 [\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] to obtain information on the area-level deprivation. This allows for IMD score, quintile, and area population for each Sub-ICB to be included. Validation of the prediction models was completed using the equivalent 2022 GP dataset.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eData Pre-processing\u003c/h2\u003e \u003cp\u003eThe variables used were IMD quintile, healthcare professional type seen in the appointment, time between booking and appointment, and appointment mode. Two additional variables were derived to indicate whether the appointment was on a weekend or holiday, which affects attendance based on the time series analysis. Missing values or entries with a \u0026lsquo;data quality\u0026rsquo; flag were removed for model training.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eAnalytical methods\u003c/h2\u003e \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e \u003ch2\u003eGeneralized Linear Models\u003c/h2\u003e \u003cp\u003eGeneralized Linear Models (GLM), such as logistic regression, are regression models [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] that can be used for classification tasks. The GLM uses a function to express the expected value as a combination of inputs, which do not need to be linear. One of the main benefits of this method is the ability to interpret regression coefficients which quantify associations between variables (a.k.a features) and the outcome. For the GP dataset, we assumed a binomial distribution for the number of missed appointments among the total number of appointments, since in our specific case, we have aggregate data with counts of appointments, weighted using the number of total appointments. Using frequency weights in a GLM model produces the same results as repeating observations by those frequencies. Training was completed considering four original dataset variables plus two additional variables to encode whether the date was a weekend or holiday. The Python module \u0026lsquo;statsmodels\u0026rsquo; was used for fitting the GLM analysis.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eExtreme Gradient Boosting\u003c/h2\u003e \u003cp\u003eExtreme gradient boosting (XGBoost) [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] is a method of gradient boosted trees, available in an open source package. The basis for this method is gradient boosting, which is a machine learning method whose goal is to find a function that approximates the output variable weights based on minimizing a loss function and makes predictions using a series of decision trees. For this model, the standard loss function used is squared error, the number of decision trees was set to 1000, and learning rate was 0.1. This method has seen much success in international machine learning competitions in a wide range of use cases, including sales prediction, event classification, and motion detection. In this case, we use the \u0026lsquo;total appointments\u0026rsquo; variable as an input since it will be assigned a weight within the algorithm. Feature importance, specifically gain, is a common metric for XGBoost and other tree-based models. Gradient boosting methods can show which variables are used as tree branches more than others, which indicates importance. This feature importance scoring is referred to as gain. A higher value of gain implies increased importance for making a prediction. Another way of assessing variable effects is by measuring the average training loss reduction gained when using a feature for splitting.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section3\"\u003e \u003ch2\u003eAssessing Prediction Performance\u003c/h2\u003e \u003cp\u003eFor accuracy, mean squared error (MSE) and mean absolute percentage error (MAPE) are used. MSE performs the squared difference between observed and predicted value for each individual observation averaged over the total. MAPE evaluate the accuracy as the ratio of the difference in observed and forecasted values (in absolute value) divided by the actual value. This is then summed for every forecasted point in time and averaged. Both metrics help quantifying errors made when using our model-based predictions compared to the observed values (smaller values the better), but the MAPE will be more sensible to errors made when the true value is small as compared to the MSE.\u003c/p\u003e \u003cp\u003eTo assess the agreement between observed and predicted probabilities of missed appointment, we also looked at the calibration by comparing the predictions obtained using both methods to the observed value for a single non-holiday weekday, varying other pre-specified appointment characteristics. These characteristics were appointment mode, IMD quintile and the number of days between booking and appointment. For investigating the trend according to a given characteristic, we had to fix the others, and the fixed values chosen were face-to-face for the appointment mode, quintile 3 for the IMD, and 8\u0026ndash;14 days for the time between booking and appointment. In other words, if we were to compare the prediction with the observed data according to IMD quintiles, we fixed the other variables to face-to-face mode and 8\u0026ndash;14 days between booking and appointment, for a single non-holiday weekday.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eDescriptive Statistical Characteristics\u003c/h2\u003e \u003cp\u003eThe analysed dataset contained 284,316,406 appointments (95.3% of the full dataset), while 13,262,722 (4.7%) appointments that were marked \u0026lsquo;unknown\u0026rsquo; or \u0026lsquo;data quality\u0026rsquo; were excluded. Further information on the data quality flag can be found on the NHS Digital website [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. This data is considered experimental, which means that their coverage and quality are not optimised. Many GP practices are not included if they use systems not eligible for inclusion in the dataset.\u003c/p\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, most appointments in the dataset were attended (95.4%) with 57.2% of all appointments taking place in person. Only 2.5% were booked well in advance (more than 28 days) compared to 45.8% same day. Most patients were seen either by a GP (51.9% of the time) or by another member of practice staff (45.0%). When looking at appointment mode, a greater proportion of missed appointments were categorized as face-to-face (78.1%). More missing appointments tend to be skewed towards lower IMD quintiles compared with attended appointments, which suggests deprivation may play a role. Missed appointments were also made normally days in advance, while the majority of attended appointments happened on the same day they were booked (47%).\u003c/p\u003e \u003cp\u003eBy plotting the percentage of daily missed appointments in 2021, we saw clear patterns in face-to-face, telephone, and video conference/online modes (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). Each peak occurs on the weekend, due to the decreased number of appointments available during these days. In video conference mode, the number of peaks in missed appointments increases in the final quarter of 2021. Less of a pattern is seen in the home visit category, with random peaks occurring over the course of the year. Autumn and winter months see an increase in the percentage of weekend missed appointments, which is not related to the total number of appointments, since there is actually an increase in appointments made in winter months.\u003c/p\u003e \u003cp\u003eWe examined trends in missed GP appointments for different sub-ICBs. Figure\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the contrast between area-level deprivation and the proportion of missed appointments by area in England. Places with high deprivation in the Northwest seem to have high rates of missing appointments. However, low deprivation areas in the Midlands and Southeast seem to also have high missing rates. Therefore, visually, we have no clear relationship between IMD score and appointment attendance. Clusters of higher missed percentages were found in areas containing large cities, such as London, Birmingham, and Manchester. These are cities that tend to be more deprived and have unequal access to medical resources. When stratifying the data based on appointment mode we find that face-to-face appointments in all areas have a higher likelihood of being missed compared to other modes. Video conference calls also have a higher percentage of missed appointments in most areas compared to home visits and telephone calls.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eMultivariable Analyses: Odds Ratios and Feature Importance\u003c/h2\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e shows the associated odds ratios of missed appointment for each of the variables. One can see a clear pattern where an increase in number of days between booking and appointment corresponds to an increase in the probability of missing the appointment. IMD status seems to influence appointment missingness, since there is decreased odds as deprivation decreases (OR 0.7). It is interesting to note that our model found that non-GP health professionals have an increased odds of having their appointment missed. If days of the week are considered individually, we find that Monday through Friday have similar percent missed rates. This is why we have considered only whether the day was a holiday or weekend. Holidays and weekends are highly associated with increased rate of missed appointments, which can be seen in the initial time series analysis as well. The next step to look at trends of missingness across different sub-ICB locations to see if certain areas are more prone to high missing rates due to medical access. For XGBoost, when measuring loss reduction gain, we find that total appointments, appointment mode, very short and very long time between booking and appointment, and whether or not it was a weekend had the most success (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eTime Series Prediction\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e shows the results of using the GLM and XGBoost models fitted to 2021 data to predict the summer seasonal results of 2022 data. This graph was obtained by averaging all the predicted values over the course of a day into a single daily value. As seen, GLM performs well with the majority of days, but underestimates weekday holiday cases. Since XGBoost does not have linear fitting, it performs better than GLM when accounting for holidays and weekends. However, due to the nonlinear fitting, the danger of overfitting increases. The only date that proves difficult for XGBoost to predict was a bank holiday that was not in the 2021 training dataset (19 Sept). Therefore, if using this as a prediction tool, it would be important to carefully label any holidays that did not occur in previous years.\u003c/p\u003e \u003cp\u003eOnce we make predictions based on the test dataset, which comes from a different year (2022), we obtain both MSE or MAPE for both methods (Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). MSE and MAPE should ideally be small and close to zero. In terms of individual predictions based on the training features, XGBoost outperforms GLM for the percentage missed appointments over the year. However, this is due in part to the high percent error for the holiday dates that GLM is unable to predict. When training and testing on weekdays only, the overall MSE and MAPE decreases for both methods.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eCalibration\u003c/h2\u003e \u003cp\u003eIn Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e, we observe the results of predicting percent missed appointments on a single weekday, 1st of March 2022. GLM and XGBoost have similar accuracy results when comparing the predictions to the actual data, but GLM obtains higher accuracy in two out of the three categories than the XGBoost model. However, days are highly variable, so it is difficult to come to a conclusion overall on which method performs better. For example, when predicting results for July 1, 2021, XGBoost outperforms GLM. Generally, GLM seems to better estimate the effect of varying appointment mode, while XGBoost has higher accuracy when predicting the effect of time between booking and appointment. Further tuning of both models is needed for predicting home visit appointment attendance, since both greatly overestimated the percent missed.\u003c/p\u003e \u003cp\u003eFrom the results from this single day, the predicted probability of a patient missing an appointment (reference values: face-to-face appointment, IMD-3 and 8\u0026ndash;14 days between booking and appointments) seems to correlate directly with the time between booking and appointment and increased deprivation. Without stratifying by appointment mode and other factors, it is impossible to know whether this is due to factors relating to IMD status. Both Face-to-face and video conference appointment modes have similar attendance rates, while telephone mode has lower predicted rates. Using this information, GP practices could send reminders in the immediate lead-up to an appointment made more than a week before for a better attendance rate. We also compare the performance of single day prediction of a model trained on all of the data versus a model trained only on weekday data. When predicting this single day using the weekday only model, the results yield slightly higher accuracies, but increased testing on other daily data would be needed to confirm this.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003ePrevious work has explored the factors in missing appointments and suggested potential ways to improve the rate of patient involvement. Many of these studies benefit from having individual-level information about each appointment that was not available for our research. However, our work uses a dataset that contains information for the whole of England, which provides insight into trends on a larger scale and on actionable health system components. We found that missed GP appointments can be predicted by appointment characteristics, such as mode, time between booking and appointment, type of day (holiday or weekend). Higher odds of missing an appointment were due to a long delay between the appointment being booked and the appointment actually taking place, and the fact that the consultation was planned with a non-GP health professional.\u003c/p\u003e \u003cp\u003eWe also estimated that deprivation affects the probability of missing an appointment, with lower probabilities of missing appointment for the least deprived areas. However, these probabilities are still non negligeable in least deprived areas, which reflects that there are other shared factors between areas (more or least deprived) that are not accounted for in this work, such as accessibility of consultations, opening hours of the GP practice, etc. The potential role of accessibility seems also supported by the lower odds of missing appointment with telephone consultation mode.\u003c/p\u003e \u003cp\u003eIn a study conducted in West Yorkshire, adult patients who missed a GP appointment and their counterparts who were present were sent a questionnaire asking about factors contributing to their decision to attend or not attend [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. Over 40% of the patients who missed appointments said they had forgotten, but others cited personal commitments or illness. This seems to agree with the results we found, as a strong predictor of missing an appointment was time between booking and appointment. Within one NHS healthcare trust, the highest predictor of unkept appointments was whether a patient had already missed one within the previous year [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]. Unfortunately, this information was not available for our particular work, but could be useful here. For a cohort analysis in 11 Scottish NHS health boards, patient data was used to analyse differences between patients who miss multiple appointments versus a few [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]. It was found that older patients (76\u0026ndash;90 years) were most likely to have a high proportion of missed appointments. More urban practices had a higher risk of missed appointments compared to rural areas which was also shown in our study. Again, these results suggest that including more details about the individual making the appointment and their interaction with the GP characteristics, would contribute to prediction success. Therefore, one limitation of our work is the lack of information on patient and practice characteristics. Further investigation is needed into the mechanism behind the increased odds of missing appointment when a non-GP health professional was being seen.\u003c/p\u003e \u003cp\u003eThe results from this and previous work suggest that practice-level modifications might be useful in preventing time and resource loss. For example, since most missed appointments are scheduled 2\u0026ndash;3 days earlier, these could be flagged as high risk and multiple reminders could be sent by text on the mobile phone. Further research is needed in meaningful ways of engaging patients who miss their scheduled appointments and ways to encourage local areas to collect appointment statistics.\u003c/p\u003e \u003cp\u003eAnother important consideration relates to deprived populations. A diabetes clinic found that their non-attending patients tended to be single parents, which supports the hypothesis that individual social factors associated with deprivation might contribute to non-attendance probability [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. Similarly, an Exeter-based GP found that low socioeconomic status and younger age were predictors of non-attendance. These findings might suggest that transport difficulties or job responsibilities might be reasons for not attending [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. One machine learning-based study found that outpatient appointments in Wales with a higher Townsend Index (more deprived) tended to have a higher \u0026ldquo;did not attend\u0026rdquo; rate [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]. Our results seem to agree with these findings as IMD status as the odds of missing an appointment increases as deprivation increases. Therefore, an implication of this work is that impoverished areas, which already tend to have lower healthcare standards, may also be losing more resources from cancelled and unattended appointments than their less deprived counterparts. However, we have not been able to investigate the role of other contextual factors (both within and outside the healthcare system). Such enhanced model will help reduce missing appointments and improve outcomes for patients. This will require future collaboration with hospitals, health care professionals, and patients [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis work showed the utility of using GLM and XGBoost to predict patient attendance in GP practices across England sub-ICB locations. This work has many strengths, including the extent of areas and deprivation levels across England that are featured in the analysis. The ability of XGBoost to predict time series trends in missing appointments was a key finding, as well as good calibration for both approaches. No approach clearly outperforms, and more work is needed in refining the model by adding additional variables or by comparing its performance to other ML methods. Newer 2022 practice-level data available on NHS digital website includes reason for attendance, which could improve the prediction model\u0026rsquo;s ability. However, this attendance data is only available at practice level. In addition, other sources of information, such as population, hospital or trust characteristics could be merged to see the influence of resources on GP attendance. Overall, the results of this work imply a large amount of NHS resource every year is lost across the whole of England due to individuals missing appointments. This is even more important for deprived areas, but it is crucial to consider that predictors in missed appointments should not be used to further prevent deprived groups from obtaining GP appointments. Instead, we should strive to improve health access for everyone by applying interventions that might decrease non-attendance. With further research into healthcare system prediction, there is an opportunity to better prepare for non-attendance in GP practices which may improve care quality and efficiency.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCRediT authorship contribution statement\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eMH: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Software, Validation, Visualization, Writing \u0026ndash; original draft, Writing \u0026ndash; review \u0026amp; editing.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSL, AE and BR : Conceptualization, Investigation, Writing \u0026ndash; review \u0026amp; editing.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAB: Conceptualization, Investigation, Methodology, Supervision, Project administration, Validation, Writing \u0026ndash; review \u0026amp; editing.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDeclaration of Competing Interest\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding information\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eInequalities in Cancer Outcome Network is funded by Cancer Research UK programme (Grant No. EPNCZS34).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData are freely publicly available on the NHS website https://digital.nhs.uk/data-and-information/publications/statistical/appointments-in-general-practice\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eA. J. Dobson and A. G. Barnett, An Introduction to Generalized Linear Models, 2018. \u003c/li\u003e\n\u003cli\u003eT. Chen and C. Guestrin, \u0026quot;XGBoost: A Scalable Tree Boosting System,\u0026quot; in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016. \u003c/li\u003e\n\u003cli\u003eNHS, \u0026quot;Missed GP appointments costing NHS millions,\u0026quot; 2019.\u003c/li\u003e\n\u003cli\u003eJ. Parsons, C. Bryce and H. Atherton, \u0026quot;Which patients miss appointments with general practice and the reasons why: a systematic review,\u0026quot; British Journal of General Practice, vol. 71, no. 707, pp. e406-e412, 2021. \u003c/li\u003e\n\u003cli\u003eJ. Lorenz and K. Hawkins, \u0026quot;Getting to Know the No-Show: Predictive Modeling of Missing a Medical Appointment,\u0026quot; SAS Proceedings, 2018. \u003c/li\u003e\n\u003cli\u003eS. Srinivas and H. Salah, \u0026quot;Consultation length and no-show prediction for improving appointment scheduling efficiency at a cardiology clinic: A data analytics approach,\u0026quot; International Journal of Medical Informatics, vol. 145, 2021. \u003c/li\u003e\n\u003cli\u003eD. Liu, W.-Y. Shin, E. Sprecher, K. Conroy, O. Santiago, G. Wachtel and M. Santillana, \u0026quot;Machine learning approaches to predicting no-shows in pediatric medical appointment,\u0026quot; NPJ Digital Medicine, vol. 5, 2022. \u003c/li\u003e\n\u003cli\u003eB. Klute, A. Homb, W. Chen and A. Stelpflug, \u0026quot;Predicting Outpatient Appointment Demand Using Machine Learning and Traditional Methods,\u0026quot; Journal of Medical Systems, 2019. \u003c/li\u003e\n\u003cli\u003eS. Rothenberg, B. Bame and E. Herskovitz, \u0026quot;Prospective Evaluation of a Machine-Learning Prediction Model for Missed Radiology Appointments,\u0026quot; Journal of Digital Imaging, vol. 35, pp. 1690-1693, 2022. \u003c/li\u003e\n\u003cli\u003eG. Fan, Z. Deng, Q. Ye and B. Wang, \u0026quot;Machine learning-based prediction models for patients no-show in online outpatient appointments,\u0026quot; Data Science and Management, vol. 2, pp. 45-52, 2021. \u003c/li\u003e\n\u003cli\u003eS. L. Harris and M. Samorani, \u0026quot;On selecting a probabilistic classifier for appointment no-show prediction,\u0026quot; Decision Support Systems, vol. 142, 2021. \u003c/li\u003e\n\u003cli\u003eL. A. Lekham, Y. Wang, E. Hey, S. S. Lam and M. T. Khasawneh, \u0026quot;A Multi-Stage predictive model for missed appointments at outpatient primary care settings serving rural areas,\u0026quot; IISE Transactions on Healthcare Systems Engineering, vol. 11, no. 2, 2021. \u003c/li\u003e\n\u003cli\u003eS. Sotudian, A. Afran, C. A. LeBedis, A. F. Rives, I. C. Paschalidis and M. D. Fishman, \u0026quot;Social determinants of health and the prediction of missed breast imaging appointments,\u0026quot; BMC Health Services Research, vol. 22, 2022. \u003c/li\u003e\n\u003cli\u003eNHS Digital, \u0026quot;Appointments in General Practice,\u0026quot; 24 November 2022. [Online]. Available: https://digital.nhs.uk/data-and-information/publications/statistical/appointments-in-general-practice.\u003c/li\u003e\n\u003cli\u003eGOV.UK, \u0026quot;English indices of deprivation 2019,\u0026quot; 26 September 2019. [Online]. Available: https://www.gov.uk/government/statistics/english-indices-of-deprivation-2019. [Accessed 2022].\u003c/li\u003e\n\u003cli\u003eN. Digital, \u0026quot;Appointments in general practice: supporting information,\u0026quot; 2023 July 26. [Online]. Available: https://digital.nhs.uk/data-and-information/publications/statistical/appointments-in-general-practice/appointments-in-general-practice-supporting-information#data-quality.\u003c/li\u003e\n\u003cli\u003eR. D. Neal, M. Hussain-Gambles, V. L. Allgar, D. A. Lawlor and O. Dempsey, \u0026quot;Reasons for and consequences of missed appointments in general practice in the UK: questionnaire survey and prospective review of medical records,\u0026quot; BMC Family Practice, 2005. \u003c/li\u003e\n\u003cli\u003eS. Philpott-Morgan, D. B. Thakrar, J. Symons, D. Ray, H. Ashrafian and A. Darzi, \u0026quot;Characterising the nationwide burden and predictors of unkept outpatient appointments in the National Health Service in England: A cohort study using a machine learning approach,\u0026quot; PLOS Medicine, 2021. \u003c/li\u003e\n\u003cli\u003eD. A. Ellis, R. McQueenie, A. McConnachie, P. Wilson and A. Williamson, \u0026quot;Demographic and practice factors predicting repeated non-attendance in primary care: a national retrospective cohort analysis,\u0026quot; The Lancet Public Health, vol. 2, no. 12, pp. e551-e559, 2017. \u003c/li\u003e\n\u003cli\u003eP. H. Dyer, C. E. Lloyd, R. J. Lancashire, S. C. Bain and A. H. Barnett, \u0026quot;Factors associated with clinic non-attendance in adults with type 1 diabetes mellitus,\u0026quot; Diabetic Medicine, vol. 15, no. 4, pp. 339-43, 1998. \u003c/li\u003e\n\u003cli\u003eW. Hamilton, A. Round and D. Sharp, \u0026quot;Patient, hospital, and general practitioner characteristics associated with non-attendance: a cohort study,\u0026quot; British Journal of General Practice, vol. 52, no. 477, 2002. \u003c/li\u003e\n\u003cli\u003eE. Incze, P. Holborn, G. Higgs and A. Ware, \u0026quot;Using machine learning tools to investigate factors associated with trends in \u0026lsquo;no-shows\u0026rsquo; in outpatient appointments,\u0026quot; Health \u0026amp; Place, vol. 67, 2021. \u003c/li\u003e\n\u003cli\u003eF. Grimm, \u0026quot;Predicting missed hospital appointments using machine learning - what are the risks?,\u0026quot; Medium, 2 August 2019. [Online].\u003c/li\u003e\n \u003c/ol\u003e"},{"header":"Tables","content":" \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDescriptive Statistics for GP Appointments in England and Wales, 2021.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eN (%)\u003c/p\u003e \u003cp\u003eTotal: 297,579,128\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eMissed Appointments (%)\u003c/p\u003e \u003cp\u003eTotal: 12,964,247\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAttended Appointments (%)\u003c/p\u003e \u003cp\u003eTotal: 284,614,881\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHealthcare Professional Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e154,300,596 (51.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3,790,637 (29.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e150,509,959 (52.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther Practice Staff\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e133,973,430 (45.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e8,876,611 (68.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e125,096,819 (44.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9,305,102(3.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e296,999 (2.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e9,008,103 (3.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAppointment Mode\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFace-to-Face\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e170,511,754 (57.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e10,128,862 (78.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e160,382,892 (56.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHome Visit\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,494,101 (0.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e81,448 (0.62)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1,412,653 (0.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTelephone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e114,574,823 (38.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2,274,779 (17.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e112,300,044 (39.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVideo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e1,479,434 (0.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e69,528 (0.54)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e1,409,906 (0.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9,519,016 (3.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e409,630 (3.2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e9,109,386 (3.2)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTime Between Booking and Appointment\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSame Day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e136,224,901 (45.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2,376,354 (18.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e133,848,547 (47.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1 Day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e27,784,950 (9.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1,121,970 (8.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e26,662,980 (9.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2 to 7 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e64,510,478 (21.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3,852,373 (29.7)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e60,658,105 (21.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e8 to 14 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e35,680,569 (12.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2,667,503 (20.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e33,013,066 (11.6)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e15 to 21 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e16,728,556 (5.6)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1,359,819 (10.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e15,368,737 (5.4)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e22 to 28 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e9,228,168 (3.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e793,801 (6.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e8,434,367 (3.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMore than 28 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7,301,308 (2.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e786,870 (6.1)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e6,514,438 (2.3)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUnknown\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e120,198 (0.04)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e5,557 (0.04)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e114,641 (0.04)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIMD 2019 Quintile\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1 \u0026ndash; most deprived\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e56,666,442 (19.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2,965,739 (22.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e53,700,703 (18.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e50,226,054 (16.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2,207,680 (17.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e48,018,374 (16.9)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e80,184,037 (26.9)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3,405,738 (26.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e76,778,299 (27.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e61,785,221 (20.8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2,532,201 (19.5)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e59,253,020 (20.8)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5 \u0026ndash; least deprived\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e48,717,374 (16.4)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1,852,889 (14.3)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e46,864,485 (16.5)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAdjusted Odds Ratio of missed GP appointment (England \u0026amp; Wales, 2021) estimated with the Generalised Linear Model (all variables detailed in the table were included as predictors in the model)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eOdds Ratio (OR)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAll Days\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eWeekdays Only\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHealthcare Professional Type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGP (reference category)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOther Practice Staff\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.669 (1.666\u0026ndash;1.672)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.754 (1.751\u0026ndash;1.756)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eAppointment Mode\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFace-to-Face (reference category)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHome Visit\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.910 (0.901\u0026ndash;0.918)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.938 (0.929\u0026ndash;0.946)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTelephone\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.502 (0.501\u0026ndash;0.503)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.528 (0.527\u0026ndash;0.529)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVideo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.913 (0.907\u0026ndash;0.920)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.928 (0.920\u0026ndash;0.936)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eTime Between Book and Appointment\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSame Day\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.497 (0.496\u0026ndash;0.499)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.516 (0.515\u0026ndash;0.517)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1 Day (reference category)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2 to 7 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.390 (1.388\u0026ndash;1.394)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.372 (1.369\u0026ndash;1.374)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e8 to 14 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.675 (1.672\u0026ndash;1.678)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.659 (1.655\u0026ndash;1.664)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e15 to 21 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.825 (1.820\u0026ndash;1.829)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.804 (1.799\u0026ndash;1.810)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e22 to 28 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.890 (1.885\u0026ndash;1.896)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.873 (1.868\u0026ndash;1.879)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMore than 28 Days\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.227 (2.221\u0026ndash;2.234)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2.238 (2.232\u0026ndash;2.246)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eIMD 2019 Quintile\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1 \u0026ndash; most deprived (reference category)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.835 (0.834\u0026ndash;0.836)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.842 (0.841\u0026ndash;0.844)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.782 (0.781\u0026ndash;0.783)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.763 (0.761\u0026ndash;0.764)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.789 (0.787\u0026ndash;0.791)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.767 (0.765\u0026ndash;0.768)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e5 \u0026ndash; least deprived\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.700 (0.699\u0026ndash;0.702)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.673 (0.672\u0026ndash;0.674)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u003cb\u003eDate Characteristics\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeekday (reference category)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHoliday\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0.973 (0.966\u0026ndash;0.977)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.00 (0.998\u0026ndash;1.01)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWeekend\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2.918 (2.895\u0026ndash;2.942)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e--\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \n\u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eMean Squared Error and Mean Absolute Percentage Error for both prediction methods, England \u0026amp; Wales, 2021\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\" style=\"width: 16.9424%;\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\" style=\"width: 37.3263%;\"\u003e\n \u003cp\u003eAll Data\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\" style=\"width: 29.1198%;\"\u003e\n \u003cp\u003eWeekdays Only\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" style=\"width: 16.9424%;\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 17.2071%;\"\u003e\n \u003cp\u003eMSE (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 20.1191%;\"\u003e\n \u003cp\u003eMAPE (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 9.7948%;\"\u003e\n \u003cp\u003eMSE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 19.325%;\"\u003e\n \u003cp\u003eMAPE(%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" style=\"width: 16.9424%;\"\u003e\n \u003cp\u003eGLM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 17.2071%;\"\u003e\n \u003cp\u003e44.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 20.1191%;\"\u003e\n \u003cp\u003e51.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 9.7948%;\"\u003e\n \u003cp\u003e29.3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 19.325%;\"\u003e\n \u003cp\u003e44.9\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\" style=\"width: 16.9424%;\"\u003e\n \u003cp\u003eXGBoost\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 17.2071%;\"\u003e\n \u003cp\u003e26.8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 20.1191%;\"\u003e\n \u003cp\u003e48.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 9.7948%;\"\u003e\n \u003cp\u003e16.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\" style=\"width: 19.325%;\"\u003e\n \u003cp\u003e40.7\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-3836849/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3836849/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eMissing general practice (GP) appointments represent an important challenge for healthcare systems. In England and Wales, reducing the number of missed appointments would benefit both the National Health Service (NHS) and the patients, avoiding delay in diagnosis and treatment.\u003c/p\u003e \u003cp\u003eSince the COVID-19 pandemic, appointment mode has shifted substantially, and many GP practices have started scheduling online appointments in place of face-to-face meetings. In this context, our aim was to build and compare prediction models for the probability of missing a GP appointment, as a function of appointment\u0026rsquo;s characteristics and the level of deprivation of the area where the GP practice is located.\u003c/p\u003e \u003cp\u003eWe examined all English GP appointments in 2021 and used two different statistical approaches for prediction: a generalized linear model (logistic regression) and a machine learning approach (Extreme gradient boosting). Predictions were further validated with 2022 data.\u003c/p\u003e \u003cp\u003eBoth approaches provided comparable predictions in term of calibration, with the advantage that results from the logistic regression can be interpreted as odds ratios. Longer time between booking and appointment plays an important role, as well as deprivation.\u003c/p\u003e \u003cp\u003eDeprived areas, which already tend to have lower healthcare standards, may also be losing more resources from cancelled and unattended appointments compared to their less deprived counterparts. Investigating the role of contextual factors behind these inequalities (both within and outside the healthcare system) would be an important step forward.\u003c/p\u003e","manuscriptTitle":"Predicting the probabilities of missed general practice appointments in England and Wales","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-10 21:54:47","doi":"10.21203/rs.3.rs-3836849/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"9a7f9b9a-de54-4dc4-b02a-e64a98dbbb10","owner":[],"postedDate":"January 10th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2024-01-14T03:29:13+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-10 21:54:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3836849","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3836849","identity":"rs-3836849","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}