Article Text
PDF
Abstract
Objective We assessed the association of early statin initiation with inpatient mortality among hospitalised COVID-19 patients.
Design, setting and participants This observational study emulated a hypothetical target trial using electronic health records data from Northwestern Medicine Health System, Illinois, 2020–2022. We included patients who were ≥40 years, admitted ≥48 hours for COVID-19 from March 2020 to August 2022 and had no evidence of statin use before admission.
Interventions Individuals who initiated any statins within 48 hours of admission were compared with individuals who did not initiate statins during this period.
Primary outcome measures Inpatient mortality at hospital days 7, 14, 21 and 28 were determined using hospital records. Risk differences between exposure groups were calculated using augmented inverse propensity weighting (AIPW) with SuperLearner.
Results A total of 8893 individuals (24.5% early statin initiators) were included. Early initiators tended to be older, male and have higher comorbidity burdens. Unadjusted day 28 mortality was higher in early initiators (6.0% vs 3.6%). Adjusted analysis showed slightly higher inpatient mortality risk at days 7 (RD: 0.5%, 95% CI: 0.2 to 0.8) and 21 (RD: 0.6%, 95% CI: 0.04 to 1.1), but not days 14 (RD: 0.4%, 95% CI: −0.03 to 0.9) and 28 (RD: 0.4%, 95% CI: −0.2 to 1.1). Sensitivity analyses using alternative modelling approaches showed no difference between groups.
Conclusions Early statin initiation was not associated with lower mortality contrasting with findings of previous observational studies. Trial emulation helped in identifying and addressing sources of bias incompletely addressed by previous work. Statin use may be indicated for other conditions but not COVID-19.
- COVID-19
- epidemiologic studies
- inpatients
- mortality
- statistics & research methods
Data availability statement
Data are available upon reasonable request. Because of the sensitive nature of the data analysed for this study, requests to access the dataset from qualified researchers trained in human subject confidentiality protocols may be sent to co-first author (OA) or (CA). Anonymised data that support the findings of this study may be made available from the investigative team in the following conditions: (1) agreement to collaborate with the study team on all publications, (2) provision of external funding for administrative and investigator time necessary for this collaboration, (3) demonstration that the external investigation team is qualified and has documented evidence of training human subjects.
This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/.
Statistics from Altmetric.com
STRENGTHS AND LIMITATIONS OF THIS STUDY
This was a retrospective study in electronic health records that utilised the target trial framework to assess the impact of early statin initiation on inpatient mortality among patients hospitalised with COVID-19; the target trial framework allowed transparency in assumptions and methods used to address sources of bias in the study design.
Nearly all prior observational studies failed to address prevalent user or immortal time bias; our analysis incorporated approaches to address these issues.
Although our results are adjusted for a rich set of confounders, including laboratory values, comorbid conditions and markers of socioeconomic status and estimated flexibly with doubly robust targeted maximum likelihood estimation with SuperLearner, our use of electronic health record data from a single health system may limit the generalisability of our findings.
Introduction
The COVID-19 pandemic has persisted for over 3 years, extending the substantial toll on hospitalisations and deaths globally.1 To date, while some effective medications have been identified to treat the symptoms and reduce the severity of disease, additional therapies continued to be studied to improve outcomes.2 3 During the early pandemic, researchers posited that statins, an easily-accessible pharmacotherapy, may improve COVID-19 patient outcomes due to their cholesterol-lowering effect, as cholesterol has been implicated in the viral replication and inflammatory processes that drive severe COVID-19 clinical syndromes.4 5 This hypothesis was first supported by a large observational study, which claimed that antecedent (ie, prior to admission) statin use was associated with improved outcomes in those hospitalised for COVID-19.6 Later, a large multisite study detected an association of statin use during admission with lower mortality risk.7 Since then, a meta-analysis of observational studies have confirmed statin-associated lower mortality risk in COVID-19 patients.8 9
Several randomised controlled trials (RCTs) concluded recently were designed to examine whether statins improve COVID-19 outcomes. The first RCT tested the effect of atorvastatin initiation on COVID-19 outcomes among intensive care unit (ICU) patients, and failed to demonstrate any benefit.10 11 Another RCT in hospitalised COVID-19 patients similarly did not observe lower mortality in the atorvastatin arm.12 A recent meta-analysis of four RCTs, including the two previously mentioned, found no significant difference between treatment and control for mortality (OR: 0.96, 95% CI: 0.61 to 1.51) or length of stay (mean difference: 0.21, 95% CI: −1.74 to 2.16).13–15 All trials may have been under powered due to small sample sizes and few mortality events. Moreover, all concluded RCTs did not include data from the Omicron variant, which may have more severely affected patients who are target statin users.
It is possible that the disagreement between the RCTs and the meta-analysis of observational studies is due to misalignment of the index visit, a phenomenon that was previously observed in studies of statins and outcomes in cancer patients. There, a meta-analysis of RCTs demonstrated no benefit (HR 1.01, 95% CI: 0.93 to 1.09), while a meta-analysis of observational studies demonstrated a substantial benefit (0.39, 0.33–0.45).16 When observational data were reanalysed using the target trial framework to emulate an RCT, the estimates were compatible with the initial meta-analysis of RCTs (1.07, 0.93 to 1.21).17
In this retrospective analysis of electronic health record (EHR) data, we used the target trial framework18 19 to assess the association of statin initiation early during inpatient stay (ie, within 48 hours of admission) on inpatient mortality risk among COVID-19 patients.
Methods
Study design and data source
Broadly, we designed a hypothetical RCT—the ‘target trial’—to examine the effect of early statin initiation during a COVID-19 inpatient stay on inpatient mortality, and conducted an analysis designed to emulate the target trial in EHR data stored by Northwestern Medicine’s Electronic Data Warehouse (NMEDW). Data concerning study eligibility were collected from 1 March 2019 to 31 August 2022, and outcome ascertainment took place between 1 March 2020 and 27 September 2022; all data were collected in the course of healthcare visits (online supplemental figure S1). Authors used data that were extracted from the EHR and cleaned by a research analyst. The first author conducted additional data cleaning and linkage to create the final analytical data sets. Details of the hypothetical trial and its emulation are outlined in table 1 (see online supplemental table S1 for full version), and key differences are described in subsequent sections.
Supplemental material
Supplemental material
Target trial specification of statin initiation on inpatient mortality among patients hospitalised with COVID-19 (short version)
Eligible population
The hypothetical trial would involve recruiting individuals aged ≥40 years with no history of prior statin use who were admitted to any Northwestern Medicine hospital with a documented positive SARS-COV-2 test (within 30 days before or up to 7 days after admission) and no contraindications for statins based on clinical or laboratory data.
In the emulation, we assumed that all individuals with no statin prescription orders in the EHR prior to admission date were statin-naïve (table 1). We assessed liver function but was unable to assess if people had previously documented statin intolerance.
Treatments and assignment
The hypothetical trial would compare two strategies: initiation (active) versus no initiation (control) of any statin at any dose during the first 48 hours of admission for COVID-19. Individuals assigned to active treatment would be allowed to discontinue at any point on advice of their clinician. This would be a pragmatic trial, so patients and clinicians would be aware of treatment assignment.
For our emulation, initiation was retrospectively assigned based on recorded clinician orders from the first 48 hours of admission. As randomisation is not possible in retrospective analyses of EHR data, individuals were assumed to be exchangeable between groups after adjustment for baseline confounders. These included (see table 1 for complete list): demographics (age, gender, race and ethnicity), site of admission, ICU and intubation status, clinical comorbidities on admission (eg, hypertension, diabetes), clinical measures (eg, heart rate, respiratory rate), laboratory measures (eg, white blood cells, C-reactive protein), medications (eg, dexamethasone) and mortality risk/illness severity scores (acute physiology score of APACHE II, sequential organ failure assessment). Comorbidities were detected using diagnosis codes (online supplemental table S2). Medications, clinical and laboratory values were measured during first 24 hours of admission. As a sensitivity analysis, models further adjusted for day 2 (hours 24–48) values of these variables, and ICU and intubation status by 48 hours of admission (time zero). Online supplemental figure S1 visualises timing of study data.
Outcomes and follow-up period
The primary outcome would be all-cause inpatient mortality within 30 days of admission confirmed by a clinician and reported to the NMEDW, conditional on surviving the initial 48 hours. The secondary outcome would be number of hospital-free days in the first month after admission, capped at 30 days. Individuals would be followed beginning 48 hours after admission, and end at the earliest of the outcome, discharge from the hospital or 30 days after admission. These were implemented in our emulation.
Statistical analysis
For the target trial, estimation of the intention to treat effects could be done by simply comparing unadjusted risks (in-hospital mortality) at prespecified times or means (hospital stay), but more commonly would be adjusted for baseline covariates for statistical efficiency.20 Here, we would estimate total average treatment effects, as differences in mortality at 7, 14, 21 and 28 days and mean differences in 30-day hospital-free time, via doubly robust AIPW, also known as one-step targeted maximum likelihood estimation (TMLE) through the ‘aipw’ and ‘tmle’ R packages.21 22 This approach draws on g-computation methods but has the property of being more robust to model misspecification: estimates are consistent if either the propensity-score (PS) model or the outcome model are correctly specified (see online supplemental methods section C for specification details of the PS and outcome models). Additionally, this approach can incorporate machine learning methods for modelling, which reduces the need to make assumptions about the functional form of the independent variables.23
Since individuals get discharged over time, another estimand we would be interested in would be is a controlled direct effect (CDE): the difference in risk of all-cause mortality had the person remained in the hospital up to a certain time (eg, outcome at day 14 had they remained in the hospital up to day 13).24 This analysis would further require specifying a model for time of discharge (censoring) and would be implemented via the ‘survTMLE’ R package.25 Our selected approach corresponds to time-to-event analyses (eg, Cox proportional hazards modelling) but provides a clearer causal interpretation.24
In our emulation, we implemented the analysis described above. Since observational data were utilised, we needed to account for covariate imbalance from non-random exposure assignment and confounding by indication; these variables are detailed in table 1 and online supplemental table S1. We specified the following collaborative learners: (mean, generalised linear model, lasso, random forest and gradient boosted trees).26–28 Several sensitivity analyses were performed. First, we repeated the analysis using parametric weighting-based approaches (inverse probability of censoring weighted pooled logistic regression for CDEs and inverse probability of treatment weighting (IPTW) for ATEs).29 This analysis assesses whether the PS component of doubly robust model reduces covariate imbalance. Second, we repeated the main analysis but limited the sample to individuals with diabetes on admission to investigate whether treatment effect heterogeneity existed in this subgroup, as suggested by previous studies.30 31 Finally, we included individuals who died or were discharged before 48 hours. To address potential immortal time bias, if individuals did not initiate statins prior to discharge, we randomly assigned them to exactly one treatment arm (see online supplemental methods section I).
All analyses were conducted using R V.≥4.10.0 with sample code in the data online supplemental file 2. Statistical significance was assessed using a 5% type-I error rate, and 95% CI were presented for all estimates; interpretation was based on the magnitude of the estimate and width of the CI in conjunction with results from statistical hypothesis tests. Since several individuals were missing laboratory data, we performed single imputation with boosted trees32 and used the imputed values to apply the laboratory criteria (see online supplemental methods section B and table S3).
Comparison to previously published observational studies
Several meta-analyses of observational studies have examined the association of statin use with mortality after COVID-19 infection.8 9 33 We were interested in assessing how published studies tackled the broad causal question were studied (ie, impact of statins on inpatient outcomes for COVID-19). Towards this goal, we identified studies included in used by three meta-analyses on statin and COVID-19 outcomes, and included those that reported a HR or OR for all-cause mortality for statin versus no statin use among COVID-19 patients followed up from time of hospital admission. We used the ‘metafor’ package to pool the study-specific estimates using a random-effects approach that relaxes the assumption that the distribution of all estimates is equal.34 35
We categorised the 22 eligible observational studies into three categories. The first group included 18 studies that compared statin users to non-users ever during admission. Most of these studies suffer from both prevalent user bias and immortal time bias. All but one failed to distinguish between initiation among statin-naïve individuals (incident users) and antecedent users who continued use during admission (prevalent users). Additionally, only three incorporated measures to address immortal time bias either through specifying an allowed initiation period or use of statistical modelling. We conducted a similar analysis in our data by classifying eligible individuals as statin users if they used statins ever during admission, they were considered users and non-users otherwise. We then estimated odds and HR for mortality at day 28 using logistic and Cox proportional hazards models adjusted for baseline covariates, like many of these papers.36 37
The second group included one study that included antecedent statin users only and then compared continuers to discontinuers. The study assigned treatment retrospectively at discharge and thus introduced immortal time bias. The third group included five studies that compared antecedent statin users who continued statin use during admission to non-users at admission, focusing on an entirely different scientific question. We did not conduct any sensitivity analyses for this group; our previous work focused on this question while here we focus on incident statin use.38
Patient and public involvement
None.
Results
Patients admitted at Northwestern Medicine with COVID-19
There were 8893 individuals included in the early initiation analysis with 2180 (24.5%) initiating statins by 48 hours of admission (online supplemental figure S2). Early initiators were slightly older and tend to be male, not vaccinated for COVID-19 and had higher comorbidity burdens (hypertension, diabetes, cardiovascular disease (CVD), chronic obstructive pulmonary disease and renal disease) (table 2, online supplemental table S4). Rates of ICU admission (14.2% vs 16.6%) and intubation (4.4% vs 4.5%) within 48 hours were comparable between the two groups (online supplemental table S5). A majority (n=8582, 96.5%) of patients tested COVID-19 positive by 48 hours of admission; there were minimal differences according to statin exposure (early initiator: 96.2% vs non-initiator: 96.6%).
Descriptive characteristics of COVID-19 patients eligible for target trial emulation, Northwestern Medicine, March 2020 to September 2022
Early initiators (n=2180) received one of the six types of statins: atorvastatin (76.8%), rosuvastatin (10.7%), pravastatin (8.7%), simvastatin (3.3%) and lovastatin or pitavastatin (4.2%). The six most common statin prescriptions were atorvastatin 40 mg (n=897, 41.1%), atorvastatin 10 mg (630, 28.9%), atorvastatin 20 mg (600, 27.5%), rosuvastatin 10 mg (198, 9.1%), pravastatin 20 mg (154, 7.1%) and atorvastatin 80 mg (133, 6.1%). Of the 2177 early initiators with dose data, 40.2% were prescribed high-intensity statins.
Association of early statin initiation with in-hospital mortality
Over the first month of admission, unadjusted mortality rates were higher in early initiators compared with non-initiators (eg, day 28: 6.0% vs 3.6%) (online supplemental table S5). The unadjusted cumulative incidence curves suggest that early initiators had higher mortality risk than non-initiators with a widening in the gap around day 14 (online supplemental figure S4)
The adjusted analysis with AIPW showed that there was a slightly higher in-hospital mortality among early initiators across all time periods but the CIs for day 14 (risk at day (RD): 0.5, 95% CI: −0.03 to 0.9) and 28 (RD: 0.4, 95% CI: −0.2 to 1.1) cross the null (table 3A). After accounting for multiple testing, day 7 results remained significant (RD: 0.5, 95% CI: 0.2 to 0.8, adjusted p-value<0.05) but day 21 did not (RD: 0.6, 95% CI: 0.04 to 1.1, adjusted p-value>0.05). No evidence of difference was observed in the sensitivity analysis using a parametric inverse propensity weighting (IPW) approach (table 3B). In sensitivity analysis which included people who were admitted for >24 but <48 hours, day 7 mortality was also higher in statin initiators, but statistical significance did not remain after accounting for multiple corrections. The subgroup analysis limited to people with DM showed null effects of statins on outcomes (online supplemental table S6). Exploratory analysis comparing low and high intensity statin users with non-users (n=8890) demonstrated no differences in mortality risk at all times assessed (online supplemental table S7).
Difference in inpatient mortality and risk between statin initiation groups, Northwestern Medicine, March 2020 to September 2022
Cumulative incidence curves of early initiators and non-users were similar and overlapping across the different analytical approaches except for gap around day 14 in the survTMLE analysis (figure 1).
Adjusted cumulative incidence of inpatient mortality, Northwestern Medicine, March 2020 to September 2022. Green lines—non-initiators, purple lines—early statin initiators. Solid lines are point estimates and dash lines are 95% CIs. Data analysis for each component are as follows. (A) main analysis comparing early initiators with never users, (B) sensitivity analysis using a parametric survival model, (C) same approach (SA) as main analysis but limited to people diagnosed with diabetes mellitus (DM) at admission and (D) same approach as main analysis but includes people admitted for >24 hours but less than 48 hours.
Outcome: hospital-free stay
Unadjusted analysis suggested that early initiators had longer median length of stay (11.5 vs 15.2 days) and shorter 30-day hospital free time (18.5 vs 14.8 days) (online supplemental table S5). On adjusted analysis with AIPW, 30-day hospital free time was found to be comparable between the two groups with slightly longer hospital free time in early initiators (average group difference: 0.08, 95% CI: −0.10 to 0.26). There were no differences even in the subgroup analysis of patients with diabetes (0.11, 95% CI: −0.26 to 0.48) or using a parametric approach (0.15, 95% CI: −0.26 to 0.56). Exploratory analysis showed that when compared with non-initiators, those who initiated high-intensity statins had longer hospital-free stay (Diffhigh-none: 0.45, 95% CI: 0.05 to 0.85) and those on low-intensity statins had comparable hospital-free stays (Difflow-none: 0.03, 95%CI: −0.34 to 0.41).
Comparison with previously published observational studies
We identified 22 studies from selected systematic reviews on statin use and COVID-19 outcomes (table 4). All studies reported lower mortality risk, but nearly all failed to address prevalent user and/or immortal time bias. Simple pooling of these studies suggested lower risk in statin users. The lone study that addressed these two issues adequately reported higher crude mortality rates among new initiators compared with statin naïve non-initiators, but lower hazard of death on adjusted analysis.39 In contrast, our findings show no difference between exposure groups (table 3; day 28 OR: 1.11, 95% CI: 0.98 to 1.26). If we ignore immortal time, 2430 individuals were ever statin initiators. Day 28 inpatient mortality risk was higher in ever versus never users (OR: 1.56, 95% CI: 1.22 to 2.00; HR: 1.25, 95% CI: 1.00 to 1.55).
Observational studies of statins and all-cause mortality among patients with COVID-19 included in recent meta-analyses
Discussion
We demonstrated that early initiation of statin during admission for COVID-19 was not associated with significant reductions in inpatient mortality or length of stay, even in individuals with pre-existing diabetes. High-intensity statin initiators had longer hospital-free stay (ie, shorter admissions) but no difference in mortality outcomes compared with non-initiators. Prior work proposed that statins improve COVID-19 outcomes through cholesterol reduction which in turn lowers inflammatory response and possibly inhibit viral replication.4 5 Even though statins are rapidly absorbed, its full effect on lipids may take several weeks to manifest.40 The lack of mortality benefits maybe due to this delay. The higher day 7 and 21 mortality in early initiators may be due to residual unadjusted confounding including confounding by indication.
Our findings were consistent with the two larger trials (INSPIRATION-S10 and RESIST12) that showed no difference in all-cause mortality between atorvastatin and placebo during ICU or hospital admission. Our unadjusted mortality rates were much lower than INSPIRATION-S but higher than RESIST (eg, 30-day mortality in statin users: 6% in our study vs 31% in INSPRITATION-S vs 3.2% in RESIST) due to differences in the study population (included non-ICU and ICU admissions), context (US vs Iran vs India) and phase of the pandemic (study period covers March 2020 to September 2022, vs July 2020 to April 2021 or July 2020 to January 2021).
While reviews found lower mortality rates in statin users, design or analytical choices may have led to unaddressed sources of bias like positivity violation (eg, statin use in <40 is rare due to practice guidelines), immortal time or incomplete adjustment for confounding.33 41–44 In our analyses, we used the target trial framework to make explicit choices that mitigate these issues. First, we identified inpatient initiators and excluded prevalent users. Second, we limited our sample to people 40 or older since in practice, people less than 40 are not likely to be on statins. This limited our generalisability but helped address positivity issues and clarified the target population. Third, we focused on early statin initiation rather than a time-varying statin initiation since adjusting for time-varying exposure was not feasible. This choice addressed immortal time bias present in prior studies. Notably, a similar study on corticosteroids and mortality in COVID-19 showed that the target trial framework can successfully recover results of randomised trials even when using observational data.45
Strengths and limitations
A strength of our analysis was the use of doubly robust techniques to better account for confounding by indication. Patients prescribed statins during hospitalisation were individuals who likely needed to be on statins prior to admission and had worse baseline health as shown by higher prevalence of comorbidities, and worse unadjusted mortality rates. In adjusted analysis, however, this difference decreased or disappeared. Known associations between comorbidities and severe COVID-19 outcomes may be driving mortality differences according to statin use during admission.46 47 We found similar findings from parametric IPW but the CIs were wider. Suggesting that, despite the computational demands, AIPW may help detect smaller differences, had they been present. Analysis ignoring immortal time yields similar findings, but the results are not interpretable.
Our study has several limitations. First, we were limited by our single health system data source in terms of variable availability and completeness during clinical care. This affected our emulation, which we documented in table 1. However, these artefacts are unlikely to affect our findings, though they may not be broadly generalisable. The data also affect generalisability of findings. Second, while we excluded known antecedent users, there is still a possibility that some early initiators were antecedent users who experienced delays in updates to active medications lists, especially if they have not been previously seen in our system. However, the majority of our sample (70%, see table 2) had an encounter with the healthcare system in the year prior to admission, indicating good participation. Third, while we were able to use various functional forms for our modelling, our results still rely on the assumption that all important confounders were measured and included in our models. Exploring this further, we obtained E-values based on risk differences ranging from 2 to 3.48 These magnitudes of association are uncommon in the literature but does not rule out unadjusted confounders especially since there is no clear mechanistic explanation for statins themselves to cause harm in people infected with COVID-19. Fourth, due to our data source, our intervention focused on early initiation rather than initiation and continuation of statins during admission (eg, initiate during first 48 hours of admission and continue until day 7 of admission). It is possible that prolonged use might translate to better outcomes, but this was difficult to emulate with our data. Relatedly, we are unable to assess the specific reasons for starting statins on each individual patient and relied on exposure modelling to mitigate bias by indication. Importantly, this model incorporates covariates that are known to influence clinical decision-making (eg, CVD history, blood pressure). Fifth, to preserve computational time, we used single rather than multiple imputation. Incorporating the additional variance from multiple imputation may widen our CIs but is not likely to affect the direction of our conclusions (ie, no benefit). Sixth, while we wanted to target admissions due to COVID-19, we are unable to identify the primary cause of admission based on available data without a resource-intensive chart review. As such, some admissions maybe for reasons other than COVID-19. Seventh, most initiators used atorvastatin. We are unable to explore any individual statin-specific effects. Finally, double robust methods have been found to perform worse if both exposure and outcome models are mis-specified.49 50 Tools used in other PS-based methods (eg, assessing balance) have limited utility of these methods and it is difficult to assess success. Instead, we conducted sensitivity analysis using parametric IPTW which showed similar results.
Conclusions
In this analysis of data from a single health system in Chicago, we found no evidence that early statin initiation reduced mortality in individuals admitted for COVID-19. Our findings back up current practice recommendations that statins should not be used to treat or mitigate the severity of COVID-19.51 Statins should be considered on an individual basis as determined by current guidelines for management of hyperlipidemia and/or CVD treatment and prevention.52
Data availability statement
Data are available upon reasonable request. Because of the sensitive nature of the data analysed for this study, requests to access the dataset from qualified researchers trained in human subject confidentiality protocols may be sent to co-first author (OA) or (CA). Anonymised data that support the findings of this study may be made available from the investigative team in the following conditions: (1) agreement to collaborate with the study team on all publications, (2) provision of external funding for administrative and investigator time necessary for this collaboration, (3) demonstration that the external investigation team is qualified and has documented evidence of training human subjects.
Ethics statements
Patient consent for publication
Ethics approval
This study received ethical approval and waiver of informed consent from the Northwestern University Institutional Review Board (IRB #: STU00212267). All methods were carried out in accordance with relevant guidelines and regulations and followed IRB protocols.
Acknowledgments
We would like to acknowledge Matthew Caputo’s assistance in curating the analytical data set for this research. A version of this work has been presented as a poster at the Society of Epidemiologic Research Conference 2022.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Contributors Substantial contribution to the conception and design: CA, OA-H, AR, LP and BT. Acquisition of data: CA and BT. Drafting of the manuscript, statistical analysis and guarantor: AR and LP. All authors (AR, OA-H, MJF, JW, BT, CA and LP) contributed to interpretation of results, contributed to the critical revision of the manuscript and gave approved the final manuscript.
Funding Research reported in this publication was supported, in part, by the National Institutes of Health's National Center for Advancing Translational Sciences, Grant Number UL1TR001422. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Competing interests LP receives funds for unrelated research from Omron Healthcare Co. Other authors have no other conflicts to declare.
Patient and public involvement Patients and/or the public were not involved in the design, or conduct, or reporting or dissemination plans of this research.
Provenance and peer review Not commissioned; externally peer reviewed.
Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.