Association of Tramadol vs Codeine Prescription Dispensation With Mortality and Other Adverse Clinical Outcomes

Key Points

Question  Is tramadol compared with codeine prescription dispensation associated with differences in the risk of mortality and other clinical outcomes?

Findings  In this retrospective cohort study that used propensity score matching and included 368 960 participants, a new prescription dispensation of tramadol, compared with codeine, was significantly associated with a higher risk of all-cause mortality (HR, 2.31), cardiovascular events (HR, 1.15), and fractures (HR, 1.50), but there was no significant difference in the risk of constipation, delirium, falls, opioid abuse/dependence, or sleep disorders.

Meaning  New prescription dispensation of tramadol, compared with codeine, was significantly associated with a higher risk of mortality, cardiovascular events, and fractures, although the findings should be interpreted cautiously, given the potential for residual confounding.

Importance  Although tramadol is increasingly used to manage chronic noncancer pain, few safety studies have compared it with other opioids.

Objective  To assess the associations of tramadol, compared with codeine, with mortality and other adverse clinical outcomes as used in outpatient settings.

Design, Setting, and Participants  Retrospective, population-based, propensity score–matched cohort study using a primary care database with routinely collected medical records and pharmacy dispensations covering more than 80% of the population of Catalonia, Spain (≈6 million people). Patients 18 years or older with 1 or more year of available data and dispensation of tramadol or codeine (2007-2017) were included and followed up to December 31, 2017.

Exposures  New prescription dispensation of tramadol or codeine (no dispensation in the previous year).

Main Outcomes and Measures  Outcomes studied were all-cause mortality, cardiovascular events, fractures, constipation, delirium, falls, opioid abuse/dependence, and sleep disorders within 1 year after the first dispensation. Absolute rate differences (ARDs) and hazard ratios (HRs) with 95% confidence intervals were calculated using cause-specific Cox models.

Results  Of the 1 093 064 patients with a tramadol or codeine dispensation during the study period (326 921 for tramadol, 762 492 for codeine, 3651 for both drugs concomitantly), a total of 368 960 patients (184 480 propensity score–matched pairs) were included after study exclusions and propensity score matching (mean age, 53.1 [SD, 16.1] years; 57.3% women). Compared with codeine, tramadol dispensation was significantly associated with a higher risk of all-cause mortality (incidence, 13.00 vs 5.61 per 1000 person-years; HR, 2.31 [95% CI, 2.08-2.56]; ARD, 7.37 [95% CI, 6.09-8.78] per 1000 person-years), cardiovascular events (incidence, 10.03 vs 8.67 per 1000 person-years; HR, 1.15 [95% CI, 1.05-1.27]; ARD, 1.36 [95% CI, 0.45-2.36] per 1000 person-years), and fractures (incidence, 12.26 vs 8.13 per 1000 person-years; HR, 1.50 [95% CI, 1.37-1.65]; ARD, 4.10 [95% CI, 3.02-5.29] per 1000 person-years). No significant difference was observed for the risk of falls, delirium, constipation, opioid abuse/dependence, or sleep disorders.

Conclusions and Relevance  In this population-based cohort study, a new prescription dispensation of tramadol, compared with codeine, was significantly associated with a higher risk of subsequent all-cause mortality, cardiovascular events, and fractures, but there was no significant difference in the risk of constipation, delirium, falls, opioid abuse/dependence, or sleep disorders. The findings should be interpreted cautiously, given the potential for residual confounding.

Opioids are used in the pharmaceutical treatment of moderate to severe cancer pain and are commonly prescribed for chronic noncancer pain when alternative therapies no longer provide sufficient relief.^1,2 Tramadol was considered to be a relatively safe opioid^3-5 and in 2013 was strongly recommended by the American Academy of Orthopaedic Surgeons for patients with symptomatic knee osteoarthritis.⁶ Data from opioid utilization studies conducted in 2019-2020 found that tramadol was the most prescribed opioid in the UK,⁷ The Netherlands,⁸ and Spain.⁹ In the US, the age-adjusted rate of drug overdose deaths involving synthetic opioids (mostly fentanyl but also tramadol) increased from 1.0 per 100 000 in 2013 to 11.4 in 2019.¹⁰

Tramadol’s common non–life-threatening adverse effects (eg, headache, itching, and nausea) are well-established. However, evidence regarding serious adverse events compared with other opioids is still scarce.^11-13 An observational study found that tramadol initiation was associated with increased risk of all-cause mortality in patients with osteoarthritis.¹⁴

Given the widespread use of tramadol to manage a broad spectrum of painful conditions, its safety profile should be comparatively assessed and its beneficial effects weighed against potential harms. This population-based cohort study compared risks of all-cause mortality, cardiovascular events, fractures, constipation, delirium, falls, opioid abuse/dependence, and sleep disorders between patients who received prescription dispensations for tramadol vs codeine. It also investigated whether the relative risk was consistent across different pain indications in adults.

This retrospective cohort study used data from the System for the Development of Research in Primary Care (SIDIAP), which comprises routinely collected anonymized electronic primary-care records (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10] codes) and sociodemographic data for a representative sample (>80%) of Catalonia, Spain. This database links with the national pharmacy claims for community pharmacy dispensations (Anatomical Therapeutic Chemical code, dispensation date, and number of packages). Previous studies have demonstrated the database’s validity and quality for research.^15-17

The local ethics committee (Comité Ético de Investigación of the Fundació Institut Universitari per a la recerca a l’Atenció Primària de Salut Jordi Gol i Gurina) approved this study (registration number P18/085). Informed consent was not required. The study protocol was registered in the European Union Post-Authorisation Studies register (EUPAS36689).

We conducted a retrospective cohort study to compare the risks associated with new prescription dispensations of tramadol vs codeine. Codeine was used as the active comparator to mitigate confounding by indication,¹⁸ as both tramadol and codeine are weak opioids with similar indication coverage and are frequently prescribed in Spain.¹⁹ “New dispensation” was defined by applying a fixed 12-month look-back period in which the patient had continuous data coverage but did not have any opioid dispensation. If a patient had multiple new dispensation episodes, only the first one was used.

All patients who were dispensed tramadol, codeine, or both between January 1, 2007, and December 31, 2017, were identified and categorized in the tramadol or codeine cohort according to the first redeemed drug. Patients with both drugs dispensed on the same day were excluded.

Cohorts were constructed by including patients who were 18 years or older, had at least 1 year of database enrollment before the first drug dispensation (index date), and had not been dispensed any opioids in the 12 months before or on entering the study. Those with a history of burn injuries, traffic crashes, major surgical procedures (amputation and joint replacement surgery), or any of the studied outcomes on or before the index date were excluded.

The study population was characterized at the index date considering the following potential confounders: sociodemographic factors (age, sex, residence region, and socioeconomic deprivation), medical conditions, Charlson Comorbidity Index, concomitant medication use (psychotropics, nonsteroidal anti-inflammatory drugs [NSAIDs], and other analgesics), and health care utilization. Socioeconomic deprivation was measured with the MEDEA (Mortalidad en áreas pequeñas Españolas y Desigualdades Socioeconómicas y Ambientales) index, which describes socioeconomic and environmental inequalities among 175 small areas of Spain. This index is divided into equal-sized quintiles, with the first quintile representing the least-deprived and last quintile representing the most socioeconomically deprived.²⁰ Health care utilization was quantified by frequency of general practice visits and hospital admissions in the past 12 months.

Health conditions expected to be linked to the opioid prescription dispensation or to be associated with any of the study outcomes were identified and included common opioid indications (cancer, back pain, neck/shoulder pain, osteoarthritis, cough, migraine), musculoskeletal diseases, chronic obstructive pulmonary disease, fibromyalgia, dyspnea, peripheral vascular disease, angina, transient ischemic attack, osteoporosis, diarrhea, rheumatoid arthritis, malabsorption disorders, pulmonary edema, diabetes, chronic liver disease, chronic kidney disease, Alzheimer disease, and Parkinson disease. eFigure 1 in the Supplement shows the study design and covariate assessment schema.²¹

Potential confounding by indication was reduced by using propensity score matching.²² For each inclusion year, propensity scores for tramadol vs codeine dispensation were calculated using a logistic regression model that included all baseline characteristics shown in the Table.

Patients with a new tramadol dispensation were matched 1:1 to those with a new codeine dispensation using caliper matching within a caliper of 0.2 times the standard error of the logit propensity score and an exact matching of inclusion year. Individuals not matched with another patient were excluded.

The study outcomes were all-cause mortality, cardiovascular events (stroke, cardiac arrhythmia, myocardial infarction, and heart failure), fracture (hip, vertebral, and other), falls, delirium, constipation, sleep disorders, and opioid abuse (“abuse” is the term used in the ICD coding)/dependence within 1 year. The ICD-10 codes for the identification of study outcomes were provided in the eAppendix 2 in the Supplement.

Baseline characteristics of the tramadol and codeine cohort participants before and after matching were compared. Missing values for living area and socioeconomic deprivation were imputed with a single data set using binary and polytomous logistic regression models, respectively. Covariate balance was assessed using the absolute standardized mean difference (ASMD). Acceptable matching was indicated by an ASMD less than 0.1.²³

For each individual, person-years of follow-up for each outcome from the index date to the first of database disenrollment, 1 year after indexing, end of the study period (December 31, 2017), or outcome occurrence was calculated. One-year cumulative incidences were computed using the Kaplan-Meier method and survival curves were plotted. Hazard ratios (HRs) were estimated using cause-specific Cox proportional hazards regression accounting for competing risk of death.²⁴ The proportional hazards assumption was tested for exposure with respect to each study outcome using the Kolmogorov supremum test and HR was reported at 3, 6, and 9 months if the assumption was violated.²⁵ Absolute rate differences (ARDs) were calculated as I₀ × (HR − 1), where I₀ is the incidence in the control group (codeine). To ensure that the overall confidence level for multiple tests is 0.95, Bonferroni-corrected confidence intervals in the main analyses are reported, which for 8 primary outcomes is [1 – (0.05/8)]*100 = 99.375%. All estimates from subgroup or sensitivity analyses are provided with regular 95% CIs.

All subsequent analyses were only conducted for outcomes that had an incidence rate higher than 10.0 per 1000 person-years or had a statistically significant difference in the main analysis. Prespecified subgroup analyses on the matched cohort with further adjustment of propensity scores in the Cox model, categorizing by age (18-39, 40-59, and ≥60 years), sex (female and male), common pain indications (cancer, cough, neurologic diseases, back pain, neck/shoulder pain, and osteoarthritis), Charlson Comorbidity Index (0, 1-2, and ≥3), and concomitant psychotropic and NSAID prescription dispensation (yes and no) were performed. The likelihood ratio test was used to examine significant multiplicative interaction effects in the Cox model. The exploratory dose-dependent analysis was described in eFigure 3 in the Supplement.

Eight sensitivity analyses were conducted. First, a multivariable Cox-proportional hazard model to estimate the HRs of study outcomes in the entire cohort.²⁶ Second, a “stratified Cox model” to control for matching structure in the matched cohort. Third, a complete-case analysis, excluding participants with missing baseline characteristic values. Fourth, to better identify new tramadol and codeine dispensations, the look-back period was extended from 1 to 3 years. Fifth, to increase drug specificity, the analysis was limited to single-formulation tramadol, excluding combinations with another nonnarcotic chemical. Sixth, tramadol may have been prescribed to relieve pain caused by an undiagnosed disease under study, such as fractures. A 3-month lag window, excluding patients with outcomes that occurred within 3 months after the index date was introduced.²⁷ Seventh, a Fine-Gray subdistribution hazard, an alternative approach to account for the competing risk of death, was used.²⁴ Eighth, a comparison of the risks during the “on-treatment” period that began with the initial prescription date and ended 30 days after the last fill would have run out was carried out. A gap of 90 days or more without a fill terminated the on-treatment period (see eFigure 2 in the Supplement for the details).

All analyses were conducted using R version 3.6.2 (R Foundation for Statistical Computing). P values were 2-sided, and P < .05 was considered significant. eAppendix 1 and 2 in the Supplement list the Anatomical Therapeutic Chemical and ICD-10 codes used.

Of the 1 093 064 patients with a tramadol or codeine dispensation between January 1, 2007, and December 31, 2017, a total of 326 921 were first dispensed tramadol, 762 492 were first dispensed codeine, and 3651 were first dispensed both drugs concomitantly and were excluded. A total of 839 489 tramadol or codeine initiators were identified by excluding 86 660 patients who were dispensed tramadol and 163 264 patients who were dispensed codeine and did not meet eligibility criteria. After propensity score matching, the matched cohort included 368 960 participants: 184 480 in the tramadol cohort and 184 480 in the codeine cohort. Figure 1 shows the participant selection process.

The Table reports the participants’ baseline characteristics before and after propensity score matching. After matching, the 2 cohorts of new users were comparable for all observed features, with no ASMD exceeding 0.1 and most not exceeding 0.01. The mean age of patients was 52.7 (SD, 16.2) years in the tramadol cohort and 53.5 (SD, 16.1) years in the codeine cohort. The prevalence of cancer was 3.2% and 3.3%, respectively, in the tramadol and codeine cohorts. The most common diagnoses were back pain (47.5% vs 48.5%), neck/shoulder pain (28.6% vs 29.5%), and osteoarthritis (15.3% vs 15.5%). The drugs most frequently dispensed were ibuprofen (34.4% vs 34.3%) and paracetamol/acetaminophen (37.1% vs 36.8%).

The proportions of residential area and socioeconomic deprivation index data missing across the entire data set were 0.04% and 27.04%, respectively.

Figure 2 shows Kaplan-Meier plots in the matched cohorts. During 1-year follow-up, participants with a prescription dispensation of tramadol had a significantly higher risk of all-cause mortality, cardiovascular disease, and fractures (1-year cumulative incidence of all-cause mortality, 12.86 [95% CI, 12.34 to 13.38] per 1000 participants in the tramadol cohort vs 5.59 [95% CI 5.25 to 5.94] per 1000 in the codeine cohort; cardiovascular disease, 9.97 [95% CI, 9.50 to 10.43] per 1000 participants in the tramadol cohort vs 8.62 [95% CI, 8.19 to 9.05] per 1000 in the codeine cohort; and fractures, 12.07 [95% CI, 11.56 to 12.57] per 1000 participants in the tramadol cohort vs 8.08 [95% CI, 7.67 to 8.50] per 1000 in the codeine cohort). Kaplan-Meier plots for the unmatched cohorts are provided in eFigure 4 in the Supplement.

Compared with codeine, tramadol prescription dispensation was significantly associated with a higher risk of mortality (13.00 vs 5.61 per 1000 person-years; HR, 2.31 [95% CI, 2.08 to 2.56]; ARD, 7.37 [95% CI, 6.09 to 8.78] per 1000 person-years), cardiovascular events (10.03 vs 8.67 per 1000 person-years; HR, 1.15 [95% CI, 1.05 to 1.27]; ARD, 1.36 [95% CI, 0.45 to 2.36] per 1000 person-years), and fractures (12.26 vs 8.13 per 1000 person-years; HR, 1.50 [95% CI, 1.37 to 1.65]; ARD, 4.10 [95% CI, 3.02 to 5.29] per 1000 person-years).

For tramadol compared with codeine prescription dispensation, there was no significant difference in the risk of constipation (6.98 vs 6.41 per 1000 person-years; HR, 1.08 [95% CI, 0.97 to 1.21]; ARD, 0.56 [95% CI, −0.17 to 1.39] per 1000 person-years), delirium (0.21 vs 0.20 per 1000 person-years; HR, 1.02 [95% CI, 0.54 to 1.93]; ARD, 0.00 [95% CI, −0.06 to 0.18] per 1000 person-years), falls (2.75 vs 2.32 per 1000 person-years; HR, 1.18 [95% CI, 0.98 to 1.42]; ARD, 0.43 [95% CI, −0.02 to 0.98] per 1000 person-years), opioid abuse/dependence (0.12 vs 0.06 per 1000 person-years; HR, 1.91 [95% CI, 0.72 to 5.08]; ARD, 0.05 [95% CI, −0.01 to 0.24] per 1000 person-years), and sleep disorders (2.22 vs 2.08 per 1000 person-years; HR, 1.06 [95% CI, 0.87 to 1.29]; ARD, 0.13 [95% CI, −0.26 to 0.62] per 1000 person-years) (Figure 3).

Because the proportional hazards assumption was violated for the mortality (P < .001) and fractures (P < .001) outcomes, the follow-up period was shortened to 3 or less, 6, and 9 months. The HR decreased gradually as time after tramadol dispensation increased (eTable 1 in the Supplement).

The increased mortality risk associated with tramadol was significantly greater among younger compared with older adults (HR, 3.14 [95% CI, 1.82 to 5.41] vs HR, 2.39 [95% CI, 2.20 to 2.60]; P < .001 for interaction) (Figure 4). Women had a significantly greater risk of cardiovascular events than men (HR, 1.32 [95% CI, 1.19 to 1.46] vs 1.03 [95% CI, 0.93 to 1.13]; P < .001 for interaction) (Figure 5). Participants with the most-prevalent comorbidities compared with the least-prevalent comorbidities had a significantly higher fracture risk (HR, 2.20 [95% CI, 1.57 to 3.09] vs 1.47 [95% CI, 1.35 to 1.59]; P = .004 for interaction) (Figure 6). Subgroup ARDs varied depending on their baseline risk for each outcome.

The sensitivity analyses (eTable 2 in the Supplement) consistently indicated that patients in the tramadol cohort experienced significantly higher risks of mortality, cardiovascular events, and fractures than patients in the codeine cohort.

In this retrospective population-based cohort study, the prescription dispensation of tramadol was significantly associated with greater risks of mortality, cardiovascular events, and fractures than the prescription dispensation of codeine. No significant difference was found in the risks of constipation, delirium, falls, opioid abuse/dependence, or sleep disorders in the year after treatment initiation. The results remained consistent and robust across subgroup and sensitivity analyses.

Musich et al²⁸ recently reported 59% increased risk of cardiovascular hospitalizations, 41% increased risk of falls/fractures, and 21% increased risk of mortality among patients with osteoarthritis 65 years or older starting tramadol, compared with nonusers. However, this comparison has limited clinical relevance, as individuals prescribed an opioid are likely to be unhealthier than those who are not. Two recent studies using the UK’s Health Improvement Network database found patients with osteoarthritis who used tramadol had a greater risk of hip fracture (HR, 1.28 [95% CI, 1.13 to 1.46])²⁹ and myocardial infarction (HR, 1.23 [95% CI, 0.95 to 1.60])³⁰ than those who used codeine. Comparing tramadol with codeine is clinically relevant, as they are both “weak” opioids (lower morphine equivalent dose) used for the second step of the World Health Organization pain ladder, such as moderate noncancer pain.

This study expanded on previous studies by comparing tramadol with codeine for various indications in all adult age groups and found that, compared with codeine, tramadol prescription dispensation was significantly associated with a higher risk of mortality, cardiovascular events, and fractures in all age groups, independent of previous comorbid conditions.

Two studies have shown results that differ from the results of the present study. Solomon et al³¹ compared 5 opioids commonly prescribed in the US in 31 375 Medicare patients. They found that codeine users had higher incidence rates of mortality compared with tramadol users (70 vs 40 per 1000 person-years), cardiovascular events (170 vs 110 per 1000 person-years), and fractures (270 vs 70 per 1000 person-years) during 180 days of follow-up. However, this study may have been subject to selection bias, because codeine users also had greater risks of these outcomes than users of oxycodone, propoxyphene, and hydrocodone, which contradicts what might be expected from physicians’ experience and the morphine milligram equivalents of the 5 drugs.

Zeng et al¹⁴ conducted a head-to-head comparison of mortality risk between tramadol, codeine, and several NSAIDs and found that tramadol initiation was associated with an increased risk of mortality compared with NSAIDs but not codeine (HR, 0.94 [95% CI, 0.83 to 1.05]). However, they used a UK-based sample of 16 922 older patients with osteoarthritis (mean age, 71 years), who differed substantially from the general population in terms of comorbidities and polypharmacy. This sample was possibly subject to depletion of susceptibility bias, in which early increased risk of acute adverse events is attenuated after a long drug exposure. The study relied on general practice prescriptions, which are more susceptible to exposure misclassification than drug dispensation records because nonadherence to opioid medications is common.³² The participants were also unlikely to truly be new users, as weak opioids used to be available over the counter in the UK.³³

This study has several limitations. First, although tramadol and codeine have been widely indicated for managing moderate to severe pain, confounding by indication could have affected the study findings. For example, codeine is often prescribed to control coughs, and a higher prevalence of cough has been observed among patients with codeine prescription dispensation (9.1%) than among patients with tramadol prescription dispensation (6.0%) before matching. However, the results remained almost unchanged when restricting the analysis to patients with or without a prior cough diagnosis, suggesting this residual indication bias is unlikely to have had a great effect. Also, tramadol may be expected to be prescribed for patients with severe illness because of a perception that it has a relatively safe profile, which could explain part of the observed associations.

Second, despite that the propensity score matching balanced the observed baseline characteristics between the 2 groups and that the subgroup analyses produced results that were highly consistent with those from the main analysis, unmeasured confounders may have nonetheless biased the estimates.

Third, the magnitude of associations between the packages of tramadol dispensation and the adverse outcomes should be interpreted cautiously. Dispensed packages of tramadol is a crude proxy for assessing dose and could instead be a reflection of anticipated pain levels, which might explain the observed association.

Fourth, this study did not include the cause of death. Given the difficulty of disentangling how much of the mortality risk was specifically related to the actual dispensation of tramadol, this estimate may be falsely elevated.

Fifth, some of the study outcomes, especially delirium and sleep disorders, may have been underestimated in routine clinical practice. Also, tramadol or codeine prescription dispensation does not necessarily indicate exposure to either drug, as the adherence is largely unknown. However, any underestimation and nonadherence are likely to have been the same in both groups, either not affecting the between-group comparison or distorting the associations toward the null.

In this population-based cohort study, a new prescription dispensation of tramadol, compared with codeine, was significantly associated with a higher risk of subsequent all-cause mortality, cardiovascular events, and fractures, but there was no significant difference in the risk of constipation, delirium, falls, opioid abuse/dependence, or sleep disorders. The findings should be interpreted cautiously, given the potential for residual confounding.

Corresponding Author: Carlen Reyes, MD, PhD, Institut Universitari d’Investigació en Atenció Primària (IDIAP Jordi Gol), Gran Via de les Corts Catalanes, 587, 08007 àtic, Barcelona, Spain (creyes@idiapjgol.info).

Accepted for Publication: August 19, 2021.

Author Contributions: Mr Xie and Dr Reyes had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Xie, Strauss, Martinez-Laguna, Carbonell-Abella, Diez-Perez, Tadrous, Reyes, Prieto-Alhambra.

Acquisition, analysis, or interpretation of data: Xie, Strauss, Martinez-Laguna, Nogues, Collins, Khalid, Delmestri, Turkiewicz, Englund, Tadrous, Reyes, Prieto-Alhambra.

Drafting of the manuscript: Xie, Carbonell-Abella, Khalid, Tadrous, Reyes.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Xie, Khalid, Turkiewicz.

Administrative, technical, or material support: Delmestri, Reyes, Prieto-Alhambra.

Supervision: Strauss, Diez-Perez, Nogues, Collins, Khalid, Delmestri, Reyes, Prieto-Alhambra.

Conflict of Interest Disclosures: Dr Strauss reported receiving grants from Amgen. Dr Martinez-Laguna reported receiving personal fees from Amgen, Italfarmaco, Ferrer, Novartis, and Rubio. Dr Carbonell-Abella reported receiving fees for lectures/classes from Amgen, Theramex, Stada, and Gebro. Dr Diez-Perez reported receiving personal fees from Amgen, Lilly, and Theramex; receiving grants from Instituto Carlos III; and owning shares of Active Life Scientific. Dr Nogues reported receiving personal fees from Amgen, UCB, and Lilly and a grant from the Foundation for Advanced Education in the Sciences. Dr Turkiewicz reported serving as associate editor for Osteoarthritis and Cartilage. Dr Englund reported receiving personal fees from Pfizer for serving as a 1-day advisory board panel member (Tanezumab; November 26, 2019). Dr Prieto-Alhambra reported receiving grant support from Les Laboratoires Servier; that his research group has received grants and advisory or speaker fees from Amgen, Astellas, AstraZeneca, Chiesi-Taylor, Johnson & Johnson, and UCB; and that Janssen, on behalf of Innovative Medicines Initiative–funded European Health Data Evidence Network and European Medical Information Framework consortiums and Synapse Management Partners, have supported training programs, open to external participants, organized by his department. No other authors reported disclosures.

Funding/Support: This work was funded by the Fundació Institut Universitari per a la recerca a l’Atenció Primària de Salut Jordi Gol i Gurina (IDIAP Jordi Gol Foundation; grant 4R18/060-1). The research was supported by the National Institute for Health Research (NIHR) Oxford Biomedical Research Centre (BRC). Dr Prieto-Alhambra is funded through an NIHR Senior Research Fellowship (grant SRF-2018-11-ST2-004). Mr Xie has been awarded a Jardine-Oxford Graduate Scholarship and a titular Clarendon Fund Scholarship outside the submitted work.

Role of the Funder/Sponsor: The funders/sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Disclaimer: The views expressed in this article are those of the authors and not necessarily those of the NHS, the NIHR, or the Department of Health.

Additional Contributions: We acknowledge English-language editing by Jennifer A. de Beyer, DPhil (Centre for Statistics in Medicine, University of Oxford). We thank the physicians and health care professionals involved in the recording of health care data in busy clinical settings in Catalonia. We also thank Chunxiao Li, BSMed, MSc (University of Cambridge), for insightful discussion. Neither Dr de Beyer nor Ms Li received any compensation for their role in the study.