HR indicates hazard ratio.
There were insufficient data to examine subgroups by angiotensin II receptor blockers, α blockers, other vasodilators, and centrally acting antihypertensives. Error bars indicate 95% CIs. ACE indicates angiotensin-converting-enzyme; CCBs, calcium channel blockers; HR, hazard ratio; and sBP, systolic blood pressure.
eAppendix. Extended Methods
eTable 1. Inclusion, Exclusion, and Study Exit Criteria
eTable 2. Propensity Score Model for the Primary Outcome (Mortality)
eTable 3. First Drug Exposure During The Study
eTable 4. Sensitivity Analyses Excluding Patients With High Estimated Cardiovascular Risk
eTable 5. Previous Blood Pressure Lowering Trials Conducted Lower Risk Populations
eFigure 1. Flow Diagram Showing Exclusion of Patient Records to Identify the Final Cohort
eFigure 2. Histogram Showing Frequency Distribution of Mean Systolic Blood Pressure Prior to the Index Date
eFigure 3. Histogram Showing Frequency Distribution of Mean Diastolic Blood Pressure Prior to the Index Date
Customize your JAMA Network experience by selecting one or more topics from the list below.
Identify all potential conflicts of interest that might be relevant to your comment.
Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.
Err on the side of full disclosure.
If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.
Not all submitted comments are published. Please see our commenting policy for details.
Sheppard JP, Stevens S, Stevens R, et al. Benefits and Harms of Antihypertensive Treatment in Low-Risk Patients With Mild Hypertension. JAMA Intern Med. 2018;178(12):1626–1634. doi:10.1001/jamainternmed.2018.4684
Is antihypertensive treatment associated with lower risk for mortality and cardiovascular disease in patients with mild hypertension?
In this study of electronic health records of 38 286 low-risk patients with mild hypertension, no evidence of an association was found between exposure to antihypertensive treatment and mortality or cardiovascular disease. There was evidence that treatment may be associated with an increased risk of adverse events, such as hypotension, syncope, and acute kidney injury.
The findings suggest that physicians should be cautious when initiating treatment in low-risk patients with mild hypertension, particularly because such an approach may affect millions of individuals with little evidence of benefit.
Evidence to support initiation of pharmacologic treatment in low-risk patients with mild hypertension is inconclusive, with previous trials underpowered to demonstrate benefit. Clinical guidelines across the world are contradictory.
To examine whether antihypertensive treatment is associated with a low risk of mortality and cardiovascular disease (CVD) in low-risk patients with mild hypertension.
Design, Setting, and Participants
In this longitudinal cohort study, data were extracted from the Clinical Practice Research Datalink, from January 1, 1998, through September 30, 2015, for patients aged 18 to 74 years who had mild hypertension (untreated blood pressure of 140/90-159/99 mm Hg) and no previous treatment. Anyone with a history of CVD or CVD risk factors was excluded. Patients exited the cohort if follow-up records became unavailable or they experienced an outcome of interest.
Prescription of antihypertensive medication. Propensity scores for likelihood of treatment were constructed using a logistic regression model. Individuals treated within 12 months of diagnosis were matched to untreated patients by propensity score using the nearest-neighbor method.
Main Outcomes and Measures
The rates of mortality, CVD, and adverse events among patients prescribed antihypertensive treatment at baseline, compared with those who were not prescribed such treatment, using Cox proportional hazards regression.
A total of 19 143 treated patients (mean [SD] age, 54.7 [11.8] years; 10 705 [55.9%] women; 10 629 [55.5%] white) were matched to 19 143 similar untreated patients (mean [SD] age, 54.9 [12.2] years; 10 631 [55.5%] female; 10 654 [55.7%] white). During a median follow-up period of 5.8 years (interquartile range, 2.6-9.0 years), no evidence of an association was found between antihypertensive treatment and mortality (hazard ratio [HR], 1.02; 95% CI, 0.88-1.17) or between antihypertensive treatment and CVD (HR, 1.09; 95% CI, 0.95-1.25). Treatment was associated with an increased risk of adverse events, including hypotension (HR, 1.69; 95% CI, 1.30-2.20; number needed to harm at 10 years [NNH10], 41), syncope (HR, 1.28; 95% CI, 1.10-1.50; NNH10, 35), electrolyte abnormalities (HR, 1.72; 95% CI, 1.12-2.65; NNH10, 111), and acute kidney injury (HR, 1.37; 95% CI, 1.00-1.88; NNH10, 91).
Conclusions and Relevance
This prespecified analysis found no evidence to support guideline recommendations that encourage initiation of treatment in patients with low-risk mild hypertension. There was evidence of an increased risk of adverse events, which suggests that physicians should exercise caution when following guidelines that generalize findings from trials conducted in high-risk individuals to those at lower risk.
High blood pressure (hypertension) is a major risk factor for cardiovascular disease (CVD),1 the leading cause of mortality worldwide.2 Hypertension is typically defined as a sustained blood pressure at or above 140/90 mm Hg taken in the clinic, and clinical guidelines recommend treatment with lifestyle or pharmacologic interventions depending on the underlying risk of CVD.3-11 Most recently, guidelines from the American College of Cardiology/American Heart Association (ACC/AHA)10 recommend that pharmacologic treatment is initiated in high-risk patients with a blood pressure of 130/80 mm Hg or higher and for all individuals with a blood pressure of 140/90 mm Hg or higher regardless of risk.
These recommendations are considered to be controversial particularly with regard to treatment of people with low CVD risk and mild hypertension (ie, sustained blood pressure of 140/90-159/99 mm Hg), for whom there is a lack of clinical trial evidence to support initiation of pharmacologic treatment.12-17 The ACC/AHA guidelines define mild, stage 1 hypertension at even lower thresholds (130/80-139/89 mm Hg); therefore, all the patients referred to in this article as having mild hypertension would now be considered to have stage 2 hypertension in the United States. These revised definitions are primarily based on findings of the Systolic Blood Pressure Intervention Trial (SPRINT),18 but although this trial included a large number of people with mild hypertension at recruitment, all participants were considered to be at high risk of CVD and 90% were already undergoing treatment. The Heart Outcomes Prevention Evaluation 3 (HOPE-3) trial19 found benefit of treatment in patients with a baseline systolic blood pressure higher than 143.5 mm Hg, but this group included participants with moderate hypertension (mean systolic blood pressure, 154 mm Hg) and intermediate risk not low risk of CVD. Meta-analyses of these and other trials demonstrated that blood pressure lowering is effective to at least 140 mm Hg systolic, but this was predominately in groups at higher cardiovascular risk.20-22
An appropriately powered study in low-risk patients is unlikely to be conducted because of the low prevalence of outcome events in this population and the unfeasibly large sample sizes required to detect a treatment effect.10,16 Therefore, the present study aimed to use routine electronic health records to examine the association between antihypertensive treatment prescriptions and all-cause mortality, CVD, and adverse events in low-risk patients with mild hypertension.
This retrospective longitudinal cohort study was conducted from January 1, 1998, to September 30, 2015, using linked data from the Clinical Practice Research Datalink (CPRD), a database of electronic health records from England. The CPRD population has previously been shown to represent the UK population.23 Detailed extended methods are available in the eAppendix in the Supplement. The study protocol was approved by CPRD’s Independent Scientific Advisory Committee in March 2016 before obtaining the data relevant to the project (protocol given in the eAppendix in the Supplement). All data are fully anonymized so consent was not required. A project summary is published on the CPRD website (https://www.cprd.com/isac).
Individual patient data were extracted from the medical records of all patients registered at general practices that contribute to the CPRD in England with linked data to the Basic Inpatient Hospital Episode Statistics and Office for National Statistics mortality register. Eligible patients were those with mild hypertension (defined as 3 consecutive blood pressure readings of 140/90-159/99 mm Hg within 12 months) and low CVD risk (eTable 1 in the Supplement). Low-risk patients were identified by excluding anyone with a history of CVD, left ventricular hypertrophy, atrial fibrillation, diabetes, chronic kidney disease, or family history of premature heart disease. When planning the study, we decided that patients’ cardiovascular risk status would be defined by comorbidities, not cardiovascular risk score, because of concerns about the amount of relevant data that might be missing in electronic health records. A total of 7720 patients (20.2%) included in the main analysis had a previous risk score recorded, and an additional 9096 (23.8%) had available risk factor information to calculate a QRISK2 score.24 It was possible to estimate a QRISK2 score for the remaining 21 050 patients older than 25 years by inserting age- and sex-standardized mean cholesterol values and Townsend scores from the Health Survey for England25 into the algorithm to replace missing data. The resulting QRISK2 scores were used to redefine the study population (further excluding patients deemed to be at high risk of CVD) and reanalyze the primary and secondary outcomes in sensitivity analyses not prespecified in the original protocol (described below).
Patients entered the study on the index date, defined as 12 months after the date of the third consecutive blood pressure reading within the range (140/90-159/99 mm Hg) occurring after the study start date (January 1, 1998). Patients exited the study when follow-up records became unavailable (ie, the date of the most recent data upload from the practice to which a given patient was registered, the date at which a given patient transfered out of a registered CPRD practice, or the date of death or specific outcome of interest) (eTable 1 in the Supplement). The last day of follow-up for those remaining in the study was September 30, 2015 (last day of follow-up in linked data).
The exposure was defined as any antihypertensive listed in the British National Formulary (code list 1 in the eAppendix in the Supplement) that was prescribed in the 12 months between hypertension diagnosis (third consecutive blood pressure reading within range) and the index date.
All-cause mortality was chosen as the primary outcome because it is accurately captured in routine health data as part of the Office for National Statistics mortality register. Secondary outcomes included the following: (1) death or hospitalization from major cardiovascular events (myocardial infarction [MI], non-MI acute coronary syndrome [ACS], stroke, heart failure, or death from CVD)18; (2) death or hospitalization from stroke, MI, non-MI ACS, heart failure, or cancer (negative control); and (3) hospitalization with suspected adverse effects to medication (hypotension, syncope, bradycardia, electrolyte abnormalities, falls, or acute kidney injury). Outcomes were captured from coded hospital admissions in the Basic Inpatient Hospital Episode Statistics, coded diagnoses in the CPRD, and death certificates from the Office for National Statistics for deaths that occurred after the index date (code list 2 in the eAppendix in the Supplement).
Data on age, sex, race/ethnicity, patient-level deprivation (Index of Multiple Deprivation), smoking status, alcohol consumption (units per week), body mass index (BMI), pretreatment blood pressure readings (in the preceding 12 months), comorbidities (rheumatoid arthritis, hypercholesterolemia [code or most recent total cholesterol value ≥290 mg/dL; to convert to millimoles per liter, multiply by 0.0259]), and all prescribed statin and antiplatelet medications were extracted from the medical records of eligible patients.
A total of 91 patients (0.001%) were missing Index of Multiple Deprivation data and were excluded from the regression analyses. The BMI was missing for 46 644 patients (42.8%) and was therefore imputed using multiple imputation. Separate imputation models were created for each outcome and included all covariates examined in each logistic regression model and the outcome event of interest, as is recommended when creating propensity score models with partially observed covariates.26 Each model was based on 20 imputations.
Analyses were conducted using Stata, versions 13.1 and 14.2 (StataCorp). Propensity scores were used to match individuals who were prescribed antihypertensive treatment (prior to the index date) to similar individuals not prescribed treatment. Variables associated with antihypertensive medication prescription were explored in a logistic regression model (Table 1). Independent variables included risk factors for cardiovascular disease, calendar year of the index date, and the general practice to which the patient was registered (Table 1). Interactions with age, BMI, smoking, and deprivation were included. Patients were matched 1:1 using the nearest neighbor method, ensuring optimal balancing of groups at the expense of a larger sample size. The χ2, Wilcoxon rank sum, and t tests were used to compare patient characteristics between groups. A 2-tailed P < .05 was considered to be statistically significant.
The validity of propensity score matching was examined using a negative control: the association of antihypertensive treatment with an outcome not known to be affected by such treatment (cancer was used in the present study). It was hypothesized that if treatment with antihypertensives had a significant association with this outcome, there was something missing in the propensity score (ie, an unmeasurable factor of propensity for treatment, such as being generally unwell or having an unhealthy lifestyle) causing an imbalance between the treatment and control groups, rather than a true treatment effect.
The efficacy of antihypertensive treatment was examined with Cox proportional hazards modeling comparing all-cause mortality among those prescribed antihypertensive treatment before the index date compared with those not prescribed treatment. Patients were analyzed in these groups regardless of whether they subsequently stopped or started treatment during follow-up. Cumulative hazard plots were produced to display the cumulative incidence of all-cause mortality in each group. Hazard ratios (HRs) were adjusted for previous cancer diagnosis, which was found to be unbalanced at baseline but was not prespecified as a covariate in the propensity score model. A post hoc decision was made to stratify the analysis by each matched pair. Numbers needed to harm were estimated for outcomes significantly associated with treatment from event rates in each group at 5 and 10 years by using the formula described by Altman and Andersen.27 Separate models were created to examine secondary end points.
Subgroup analyses were conducted to examine the association between antihypertensive treatment and mortality or CVD, stratified by age (±65 years), sex, and antihypertensive drug class. Post hoc subgroup analyses examined the association between treatment and outcomes, stratified by baseline systolic blood pressure (±150 mm Hg). Post hoc sensitivity analyses were conducted using estimated cardiovascular risk scores by including patients’ cardiovascular risk score in each propensity score model, matching and rerunning the main analysis, and excluding anyone with a risk score of 20% or higher and then including remaining patients’ cardiovascular risk score in each propensity score model and matching and rerunning the main analysis. Additional post hoc sensitivity analyses were undertaken to examine the association between treatment and mortality by using standard multivariate adjustment instead of propensity score matching.
A total of 108 844 patients were potentially eligible for inclusion in the analysis, including 19 143 patients prescribed treatment in the 12 months before the index date (eFigure 1 in the Supplement). A total of 19 143 treated patients (mean [SD] age, 54.7 [11.8] years; 10 705 [55.9%] women; 10 629 [55.5%] white) were matched to 19 143 similar untreated patients (mean [SD] age, 54.9 [12.2] years; 10 631 [55.5%] female; 10 654 [55.7%] white), giving a total sample population for the main analysis of 38 286 patients followed up for a median of 5.8 years (interquartile range, 2.6-9.0 years). The mean (SD) blood pressure before initiation of treatment was 146/89 (6/5) mm Hg (eTable 2 and eAppendix in the Supplement). There were statistically but not clinically significant differences between the control and treatment groups in pretreatment mean (SD) diastolic blood pressure (88.5 [5.2] vs 88.7 [5.6] mm Hg; P = .002), cardiovascular risk score (8.1% [6.6%] vs 7.9% [6.6%]; P = .008), and alcohol consumption (13.0 [14.9] vs 12.1 [15.0] units per week; P = .001) (eTable 2 and eAppendix in the Supplement). A total of 7958 patients (41.6%) in the control group were prescribed an antihypertensive drug at some point during follow-up (eTable 3 and eAppendix in the Supplement), but total treatment duration among these patients was less than a third of that in the exposed group (34 571 vs 104 695 treatment years).
A total of 1641 deaths were observed across the groups during the follow-up period. Overall mortality was 4.08% (95% CI, 3.80%-4.37%) in the control group and 4.49% (95% CI, 4.20%-4.80%) in the treatment group, a risk difference of 0.41% (95% CI, 0.02%-0.85%). No significant difference was found between groups in time to death (HR, 1.02; 95% CI, 0.88-1.17; P = .81) (Table 2 and Figure 1).
No significant associations were observed between antihypertensive treatment and CVD (HR, 1.09; 95% CI, 0.96-1.25). Similarly there were no associations with stroke, MI, heart failure, or non-MI acute ACS (Table 2 and Figure 1). There was a significant association between baseline antihypertensive treatment exposure and time to adverse events, including hypotension (HR, 1.69; 95% CI, 1.30-2.20; P < .001), syncope (HR, 1.28; 95% CI, 1.10-1.50; P = .002), electrolyte abnormalities (HR, 1.72; 95% CI, 1.12-2.65; P = .01), and acute kidney injury (HR, 1.37; 95% CI, 1.00-1.88; P = .048) but not with falls or bradycardia (Table 2 and Figure 2). Numbers needed to harm for treatment prescription were as high as 580 (95% CI, 253-361) at 5 years and 111 (95% CI, 49-687) at 10 years for electrolyte abnormalities and as low as 135 (95% CI, 77-385) at 5 years and 35 (95% CI, 20-100) at 10 years for syncope. Numbers needed to harm at 10 years were 41 (95% CI, 24-93) for hypotension and 91 (95% CI, 39-14 552) for acute kidney injury (Table 2). Baseline treatment exposure was not associated with the negative control (time to cancer: HR, 1.01; 95% CI, 0.92-1.11; P = .79).
No evidence of an association was observed between baseline antihypertensive treatment and mortality or CVD by age, systolic blood pressure, or antihypertensive drug class (Figure 3). Sensitivity analyses adjusting the propensity score model for baseline cardiovascular risk score revealed a significant association between antihypertensive treatment and non-MI ACS (HR, 0.54; 95% CI, 0.33-0.89; P = .02), whereas the associations between treatment and electrolyte abnormalities (HR, 1.38; 95% CI, 0.93-2.05; P = .11) and acute kidney injury were no longer significant (HR, 1.15; 95% CI, 0.85-1.58; P = .37) (eTable 4 in the Supplement). Adjustment and exclusion of individuals estimated to be at high risk of CVD (>20% risk) produced similar results to the primary analysis except for the association between treatment and acute kidney injury, which was no longer significant (HR, 1.32; 95% CI, 0.93-1.89; P = .12). Analysis of the data using multivariate adjustment, rather than propensity score matching, showed a significant association between treatment and mortality, with smaller CIs because of the larger sample size available (multivariate adjustment: HR, 1.10; 95% CI, 1.02-1.19; propensity score matching: HR, 1.02; 95% CI, 0.88-1.17).
The present study examined electronic health records from 38 286 low-risk patients with mild hypertension and compared rates of mortality and CVD between patients prescribed treatment and those not prescribed treatment for a median follow-up period of 5.8 years. No evidence of an association was found between baseline exposure to antihypertensive treatment and mortality or CVD. There was evidence to suggest that baseline treatment exposure may be associated with an increased risk of adverse events, with a number needed to harm after 5 years of treatment of 135 for syncopal outcomes. This finding does not seem particularly important in terms of number needed to harm but, in the context of little evidence of benefit, suggests that physicians should be cautious when initiating new treatment in this population, particularly because such an approach may affect millions of individuals.17,28
A number of previous trials have examined the efficacy of blood pressure–lowering treatment among patients with mild hypertension but predominately focused on higher-risk individuals.20-22,29 Trials examining lower-risk populations are summarized in eTable 5 in the Supplement.19,30-34 In trials that found a benefit with treatment, it can be argued that participants were not truly low risk as defined in clinical guidelines,30 and some trials included patients with moderate hypertension (systolic blood pressure ≥160 mm Hg); thus, their findings are not directly relevant here.19,32 Studies that examined a relevant population found no associations between treatment and cardiovascular events,31,33,34 consistent with the findings of the present study.
A Cochrane review by Diao and colleagues29 examined 8912 patients from 4 clinical trials and found no significant reduction in mortality, coronary artery disease, stroke, or total cardiovascular events with treatment. However, the authors of that review and subsequent commentators12,29 pointed to a lack of power in previous trials and meta-analyses to show significant results. In contrast, the current study was sufficiently powered to detect a treatment association but failed to find one.
The meta-analysis by Brunström and Carlberg21 showed benefit of treatment at lower blood pressures in patients with a history of CVD and higher-risk primary prevention patients. The present study found no evidence of benefit with treatment in lower-risk populations.
The present data provide no evidence to suggest that new ACC/AHA guidelines10 will reduce CVD events in low-risk patients with mild hypertension. Even in sensitivity analyses adjusting the propensity score for previous or imputed risk score, the observed treatment benefit for cardiovascular outcomes was minimal, with only non-MI ACS associated with a significant risk reduction with treatment. Furthermore, we found that long-term antihypertensive treatment in clinical practice was associated with harm attributable to adverse events, such as hypotension, syncope, electrolyte abnormalities, and acute kidney injury, although electrolyte abnormalities and acute kidney injury were sensitive to the definition of high risk used in the sensitivity analyses. Physicians should therefore be cautious when initiating new treatment in this population, and patients should be made aware of the limited evidence of efficacy for treatment in low-risk individuals. These findings may be particularly relevant for younger patients with mild hypertension, because as these individuals age, they are likely to develop higher risk and moderate or severe hypertension, for which the benefits of treatment are more established.20-22,29
This nationally representative23 observational cohort study with a prespecified analysis plan is the largest study, to our knowledge, to examine the association between antihypertensive treatment and mortality among patients with low-risk mild hypertension. Despite this, CIs for estimates of benefit and harm outcomes were relatively wide; therefore, a larger study would be required to provide more precise results. Crossover between treatment groups was observed in the study, reflecting the observational nature of the data. Those in the treatment group were exposed to 3 times as many years of treatment than those in the control group; thus, if crossover masked an association with treatment, such an association would have been small.
Fewer events occurred for assessment of secondary outcomes, which may have been affected by inadequate documentation in the electronic health records studied. Linked data were used, which reduces the consequences of this limitation35; however, this strategy is unlikely to mitigate them completely, particularly for outcomes such as falls, which may not lead to hospitalization or reporting to a primary care physician. Arguably, events not leading to contact with medical services are less likely to be important to an individual. Assessment of secondary outcomes may also have been affected by ascertainment bias (ie, people undergoing treatment are more likely to report having adverse events), although the risk of this bias was minimized by limiting the end point to those who were hospitalized for a given event. Blood pressure differences at follow-up were not examined because monitoring strategies are likely to be dependent on whether people are undergoing antihypertensive therapy, giving potentially misleading results.
Propensity scores were used in the present analysis to balance measured confounders at the index date. Because these are nonrandomized, observational data, the results may still be biased because of unmeasured residual confounding. Cancer prevalence was found to be higher in the treatment group at baseline, and although this was adjusted for in the main analysis, we cannot rule out the possibility that other confounders existed, causing our treatment group to be higher risk than those in the control group; this scenario might explain the lack of evidence of treatment benefit. There may also have been bias in recording of risk factors (used in the propensity score model), although because people undergoing treatment are more likely to have complete data, this would likely have led to controls with higher risk than recorded, which would have favored those undergoing treatment (ie, made treatment appear to be more beneficial). Small absolute differences were observed in diastolic blood pressure, estimated cardiovascular risk, and alcohol consumption between groups at baseline, which were statistically significant, as might be expected given the large sample size, but not clinically significant (ie, differences of <0.2 mm Hg in diastolic blood pressure, <0.2% risk, and <1 unit per week of alcohol).
The current study made many comparisons without adjustment for multiple testing; therefore, caution is required when interpreting the results. In particular, some subgroup analyses suggested a possible association between treatment and CVD in women and those taking angiotensin-converting enzyme inhibitors, but these findings should be interpreted carefully because no significant interaction effects were observed. Patients included in the study cohort were followed up for a median of 5.8 years. Although this is comparable or longer than most previous trials,18,21 it is possible that the benefits of treatment take longer to manifest in this low-risk population and therefore may have become more evident had data been available to follow up patients for longer. Data for BMI were missing for 42% of patients available for analysis. Because BMI was considered to be an important potential variable associated with treatment and unlikely to be missing not at random, multiple imputation was deemed to be appropriate to avoid significant loss of data in a complete case analysis.
Exposure to antihypertensive treatment was based on prescriptions issued by a physician in primary care, but it was not possible to ascertain whether this prescription was subsequently filled or whether patients actually took the treatment as prescribed. Finally, subgroup analyses were undertaken with the sample cohort used in the primary analysis, and patients were not rematched based on propensity score, meaning that there was a small imbalance in the total numbers compared in each subgroup for age, sex, and systolic blood pressure.
These observational data provide no evidence that antihypertensive treatment is associated with reduced mortality or rates of CVD among low-risk patients with mild hypertension. Such data may be subject to bias from unmeasured confounding but suggest that caution should be exercised when considering treatment in this population.
Accepted for Publication: July 18, 2018.
Corresponding Author: James P. Sheppard, PhD, Nuffield Department of Primary Care Health Sciences, University of Oxford, Woodstock Road, Oxford, OX2 6GG, United Kingdom (email@example.com).
Published Online: October 29, 2018. doi:10.1001/jamainternmed.2018.4684
Author Contributions: Dr Sheppard had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: All authors.
Acquisition, analysis, or interpretation of data: Sheppard, S. Stevens, R. Stevens, Mant, Hobbs, McManus.
Drafting of the manuscript: Sheppard, Martin, McManus.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Sheppard, S. Stevens, R. Stevens.
Obtained funding: Sheppard, Hobbs, McManus.
Administrative, technical, or material support: Sheppard, S. Stevens.
Supervision: R. Stevens, Martin, Hobbs, McManus.
Conflict of Interest Disclosures: Dr R. Stevens is a member of the Clinical Practice Research Datalink's Independent Scientific Advisory Committee but was not involved in the approval of this study. No other disclosures were reported.
Funding/Support: This work was funded by Medical Research Council (MRC) Strategic Skills Postdoctoral Fellowship MR/K022032/1 (Dr Sheppard), a National Institute for Health Research (NIHR) professorship (Dr Sheppard and Mr McManus), and grant NIHR-RP-R2-12-015 from the NIHR (Mr McManus). Dr Sheppard receives funding from the NIHR Collaboration for Leadership in Applied Health Research and Care Oxford at Oxford Health National Health Service Foundation Trust and the NIHR School for Primary Care Research (SPCR). Mr Hobbs received support from the NIHR as director of the NIHR SPCR, director of the NIHR Collaboration for Leadership in Applied Health Research and Care Oxford, theme leader of the NIHR Oxford Biomedical Research Centre, and member of the NIHR Oxford Diagnostic Evidence Cooperative and from Harris Manchester College.
Role of the Funder/Sponsor: The funding sources 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 and opinions expressed are those of the authors and do not necessarily reflect those of the MRC, NHS, NIHR, or the UK Department of Health.
Additional Contributions: Blanca Gallego Luxan, PhD, reviewed and commented on the study protocol. She was not compensated for her work.
Create a personal account or sign in to: