Analyzing partially missing confounder information in comparative effectiveness and safety research of therapeutics

A Principled Approach to Characterize and Analyze Partially Observed Confounder Data from Electronic Health Records

Objective

Partially observed confounder data pose challenges to the statistical analysis of electronic health records (EHR) and systematic assessments of potentially underlying missingness mechanisms are lacking. We aimed to provide a principled approach to empirically characterize missing data processes and investigate performance of analytic methods.

Methods

Three empirical sub-cohorts of diabetic SGLT2 or DPP4-inhibitor initiators with complete information on HbA1c, BMI and smoking as confounders of interest (COI) formed the basis of data simulation under a plasmode framework. A true null treatment effect, including the COI in the outcome generation model, and four missingness mechanisms for the COI were simulated: completely at random (MCAR), at random (MAR), and two not at random (MNAR) mechanisms, where missingness was dependent on an unmeasured confounder and on the value of the COI itself. We evaluated the ability of three groups of diagnostics to differentiate between mechanisms: 1)-differences in characteristics between patients with or without the observed COI (using averaged standardized mean differences [ASMD]), 2)-predictive ability of the missingness indicator based on observed covariates, and 3)-association of the missingness indicator with the outcome. We then compared analytic methods including "complete case", inverse probability weighting, single and multiple imputation in their ability to recover true treatment effects.

Results

The diagnostics successfully identified characteristic patterns of simulated missingness mechanisms. For MAR, but not MCAR, the patient characteristics showed substantial differences (median ASMD 0.20 vs 0.05) and consequently, discrimination of the prediction models for missingness was also higher (0.59 vs 0.50). For MNAR, but not MAR or MCAR, missingness was significantly associated with the outcome even in models adjusting for other observed covariates. Comparing analytic methods, multiple imputation using a random forest algorithm resulted in the lowest root-mean-squared-error.

Conclusion

Principled diagnostics provided reliable insights into missingness mechanisms. When assumptions allow, multiple imputation with nonparametric models could help reduce bias.

EAT-Lancet Healthy Reference Diet score and diabetes incidence in a cohort of Mexican women

To improve the health of our planet and develop sustainable food policies, it is important to understand the health impact of a diet pattern that considers planetary and population health. We used data from the Mexican Teachers' Cohort (MTC) to estimate the association between the EAT-Lancet healthy reference diet (EAT-HRD) and type 2 diabetes (T2D) incidence. We included 74,671 women aged ≥25 years, free of T2D at baseline. A validated food frequency questionnaire (FFQ) was used to assess dietary intake. We created an EAT-HRD score based on 15 food groups recommended by the EAT-Lancet Commission (range from 0 to 15). T2D cases were identified through self-report and cross-linkage with clinical and administrative databases. We used Cox proportional hazards models to estimate the association between categories of the EAT-HRD score with T2D incidence. During a median follow-up of 2.16 y (IQR 1.8-4.3 y), we identified 3241 T2D incident cases. The median EAT-HRD score was 6 (IQR 5-7). In multivariable analyses, when comparing extreme categories, higher adherence to the EAT-HRD score was associated with lower T2D incidence (HR 0.90; 95% CI 0.75, 1.10), yet, the estimation was imprecise. Compared to those who did not meet the EAT-HRD recommendations, adhering to the red meat, legumes, and fish recommendations was associated with lower T2D incidence. Meeting the recommendation of dairy and added sugars was associated with an increased incidence of T2D. Higher adherence to a diet designed to promote environmental and human health may help prevent T2D incidence among women in a middle-income country.

Real World Outcomes in Patients With Metastatic, Castration-Resistant Prostate Cancer Treated With Radium-223 in Routine Clinical Practice in Sweden

AIM

Estimate the effect of Radium-223 (Ra-223) on the incidence of bone fractures, prostate cancer death, and all-cause death compared with other standard treatments for metastatic, castration-resistant prostate cancer (mCRPC).

METHODS

Using a cohort design, we estimated the effect of Ra-223 on the risk of bone fractures, all-cause and prostate cancer-specific mortality across different lines of treatment for mCRPC using Prostate Cancer data Base Sweden (2013-2018). The comparator group comprised other standard treatments for mCRPC. We used 36-month risk differences and hazard ratios (HRs) as effect estimates.

RESULTS

The number of eligible individuals was 635, 453, 262, and 84 for the first-, second-, third-, and fourth-line cohorts, respectively. When compared Ra-223 to other standard treatments, the difference in the 36-month risk of fracture was 6% (95% confidence interval [CI], -7% to 18%) in the first-line cohort (n = 635) and 8% (95% CI, -7% to 18%) in the second-line cohort (n = 453). The number of fractures in the third-/fourth-line cohorts was too small for an adjusted comparison. The difference in 36-month mortality was higher in the first-line cohort 13% (95% CI, -3% to 31%), but lower in the second- and third-/fourth-line cohorts-8% (95% CI, -23% to 7%) and -14% (95% CI, -21% to 16%) respectively. Most deaths were due to prostate cancer.

CONCLUSION

Results suggest that the difference in the risk of fractures is small, if any. A difference in the risk of mortality may be present in first-line treatment, but a decreased risk of mortality was observed in second and later lines of treatment. The results on mortality need to be considered in the context of potential unmeasured or residual confounding.

Choice of imputation method for missing metastatic status affected estimates of metastatic prostate cancer incidence

OBJECTIVES

To study how handling missing data on M stage in a clinical cancer register affects estimates of incidence of metastatic prostate cancer.

STUDY DESIGN AND SETTING

Estimates of age-standardized incidence of metastatic prostate cancer were obtained by the use of data in a population-based clinical cancer register in Sweden and using four methods for imputation of missing M stage. Adjusted survival was used to compare men with known and imputed M stage.

RESULTS

The proportion of men with missing M stage was high (66%) and varied according to the risk group and over calendar time. The estimated incidence of metastatic disease varied depending on imputation method, with all methods indicating a decreasing incidence over time. A combination of deterministic imputation (DI) and multiple imputation (MI) produced adjusted survival curves for men with imputed M stage that best resembled the survival for men with known M stage.

CONCLUSIONS

Plausible estimates of incidence of metastatic prostate cancer in clinical cancer registers can be obtained by the use of a combination of DI of missing M stage and MI.

Avocado consumption is associated with a reduction in hypertension incidence in Mexican women

Avocado is a fruit rich in dietary fibre, potassium, Mg, mono and PUFA and bioactive phytochemicals, which are nutritional components that have been associated with cardiovascular health. Yet, despite the boom in avocado consumption, we lack evidence on its association with CVD risk in the general population. To estimate the prospective association between avocado consumption and incident hypertension in Mexican women, we estimated the association in participants from the Mexican Teachers’ Cohort who were ≥ 25 years, free of hypertension, CVD and cancer at baseline (n 67 383). We assessed baseline avocado consumption with a semi-quantitative FFQ (never to six or more times per week). Incident hypertension cases were identified if participants self-reported a diagnosis and receiving treatment. To assess the relation between categories of avocado consumption (lowest as reference) and incident hypertension, we estimated incidence rate ratios (IRR) and 95 % CI using Poisson regression models and adjusting for confounding. We identified 4002 incident cases of hypertension during a total of 158 706 person-years for a median follow-up of 2·2 years. The incidence rate of hypertension was 25·1 cases per 1000 person-years. Median avocado consumption was 1·0 (interquartile range: 0·23, 1·0) serving per week (half an avocado). After adjustment for confounding, consuming 5 + servings per week of avocado was associated with a 17 % decrease in the rate of hypertension, compared with non- or low consumers (IRR = 0·83; 95 % CI: 0·70, 0·99; Ptrend = 0·01). Frequent consumption of avocado was associated with a lower incidence of hypertension.

Ultraprocessed beverages and processed meats increase the incidence of hypertension in Mexican women

Higher intake of ultraprocessed foods (UPF), which have undergone multiple processes and have poor nutrient quality, is associated with higher incidence of non-communicable diseases. Yet, its association with hypertension has scarcely been studied, especially in low- and middle-income countries (LMIC). We aimed to estimate the associations between consumption of UPF (total, liquid and solid) and UPF subgroups and incident hypertension in a prospective cohort study. We used data from the Mexican Teachers' Cohort including 64 934 disease-free women aged ≥25 years at baseline. We assessed baseline usual dietary intake using a validated FFQ, and each item was categorised according to NOVA, a degree of food processing classification system. UPF and UPF subgroups were categorised according to the distribution of their contribution to total energy intake. Hypertension was self-reported. We estimated incidence rate ratios (IRR) and their 95 % CI. During a median follow-up of 2·2 years, we identified 3752 incident cases of hypertension. Mean contribution of UPF to total energy intake was 29·8 (SD 9·4) % energy (23·4 (SD 8·9) % solid, 6·4 (SD 4·8) % liquid). Comparing extreme categories showed that higher total and solid UPF consumptions were not associated with incident hypertension (IRR 0·96, 95 % CI 0·79, 1·16; IRR 0·91, 95 % CI 0·82, 1·01, respectively). However, liquid UPF and processed meats were associated with increased hypertension (IRR 1·32, 95 % CI 1·10, 1·65; IRR 1·17, 95 % CI 1·01, 1·36, respectively). Addressing intake of liquid UPF and processed meats may help in managing hypertension in LMIC.

Methods for external control groups for single arm trials or long-term uncontrolled extensions to randomized clinical trials

Purpose

Clinical trials compare outcomes among patients receiving study treatment with comparators drawn from the same source. These internal controls are missing in single arm trials and from long‐term extensions (LTE) of trials including only the treatment arm. An external control group derived from a different setting is then required to assess safety or effectiveness.

Methods

We present examples of external control groups that demonstrate some of the issues that arise and make recommendations to address them through careful assessment of the data source fitness for use, design, and analysis steps.

Results

Inclusion and exclusion criteria and context that produce a trial population may result in trial patients with different clinical characteristics than are present in an external comparison group. If these differences affect the risk of outcomes, then a comparison of outcome occurrence will be confounded. Further, patients who continue into LTE may differ from those initially entering the trial due to treatment effects. Application of appropriate methods is needed to make valid inferences when such treatment or selection effects are present.

Outcome measures in a trial may be ascertained and defined differently from what can be obtained in an external comparison group. Differences in sensitivity and specificity for identification or measurement of study outcomes leads to information bias that can also invalidate inferences.

Conclusion

This review concentrates on threats to the valid use of external control groups both in the scenarios of single arm trials and LTE of randomized controlled trials, along with methodological approaches to mitigate them.

Risk of Myocardial Infarction Among New Users of Calcium Supplements Alone or Combined With Vitamin D: A Population-Based Case-Control Study

A population‐based case‐control study was conducted to evaluate the risk of acute myocardial infarction among new users of calcium supplements either in monotherapy (CaM) or in combination with vitamin D (CaD). A total of 23,025 cases and 114,851 controls randomly sampled from the underlying cohort and matched with cases by age, sex, and index date were included. New users of CaM and CaD were categorized as current users, recent users, past users, and nonusers. We computed adjusted odds ratios (AORs) and their 95% confidence intervals (CIs) among current users as compared with nonusers through a conditional logistic regression. No increased risk was associated with CaM overall (59 cases (0.26%) and 273 controls (0.24%); AOR = 0.80; 95% CI 0.59–1.09), nor was it found in any of the conditions examined. Instead, the use of CaD was associated with a decreased risk (275 cases (1.19%) and 1,160 controls (1.45%); AOR = 0.78; 95% CI 0.67–0.90), dose and duration‐dependent, and particularly evident in patients with a high cardiovascular risk (AOR = 0.59; 95% CI 0.43–0.81).

Reproducing Protocol-Based Studies Using Parameterizable Tools-Comparison of Analytic Approaches Used by Two Medical Product Surveillance Networks

The US Sentinel System and the Canadian Network for Observational Drug Effect Studies (CNODES) are two medical product safety surveillance networks. Using Sentinel's preprogrammed, parameterizable analytic tools, we reproduced two protocol-based studies conducted by CNODES to assess the risks of acute pancreatitis and heart failure (HF) associated with the use of incretin-based drugs, compared with use of ≥ 2 oral hypoglycemic agents. Results from the replication new-user cohort analyses aligned with those from the CNODES nested case-control studies. The adjusted hazard ratios were 0.95 (0.81-1.12; vs. 1.03 (0.87-1.22) in CNODES) for acute pancreatitis and 0.91 (0.84-1.00; vs. 0.82 (0.67-1.00) in CNODES) for HF among patients without HF history. The CNODES's common protocol approach allows studies tailored to specific safety questions, whereas the Sentinel's common data model plus pretested program approach enables more rapid analysis. Despite these differences, it is possible to obtain comparable results using both approaches.

Early Menstrual Factors Are Associated with Adulthood Cardio-Metabolic Health in a Survey of Mexican Teachers

Objectives To evaluate whether age at menarche and time to menstrual regularity were related to cardio-metabolic risk factors in Mexican women. Methods The study population comprised 54,921 women from the 2008-2010 wave of the Mexican Teacher's Cohort. A modified Poisson approach was used; exposures were age at menarche and time to menstrual regularity (< 1 year vs. ≥1 year), and outcomes were prevalent obesity, type 2 diabetes, high blood pressure, and high cholesterol. Results Mean (SD) age of women was 42.1 (7.6) years, and mean (SD) menarcheal age was 12.5 (1.5) years. Compared to women with menarche age 13 years, those with menarche < 9 years had a 65% (95% CI 43-90%); 27% (95% CI 4-55%); and 23% (95% CI 1-49%) higher prevalence of obesity, high blood pressure, and high cholesterol, respectively. For diabetes, there was a U-shaped association; compared to menarche age 13 years, those with menarche < 9 years had an 89% higher prevalence of diabetes (95% CI 39-156%), and those with menarche ≥ 17 years had a 65% higher prevalence (95% CI 16-134%). Among women with regular cycles (n = 43,113), a longer time to menstrual regularity was associated with diabetes (PR = 1.11 with 95% CI 1.02-1.22), high blood pressure (PR = 1.11 with 95% CI 1.06-1.17), and high cholesterol (PR = 1.09 with 95% CI 1.04-1.14). Conclusions for practice Mexican women with earlier and later ages at menarche and/or longer time to menstrual regularity may have higher risk of cardio-metabolic disease in adulthood.

Sugar-Sweetened Soda Consumption Increases Diabetes Risk Among Mexican Women

BACKGROUND

Epidemiological evidence supports an association between sugar-sweetened soda consumption and diabetes. However, evidence regarding this association is limited in countries that have recently undergone a nutritional transition.

OBJECTIVE

We estimated the association between sugar-sweetened soda consumption and incident diabetes. We also determined if the association between sugar-sweetened soda and diabetes differs as a result of early life factors and potential genetic susceptibility.

METHODS

We used data from the Mexican Teachers' Cohort including 72,667 women aged ≥25 y, free of diabetes, cardiovascular disease, and cancer at baseline. We assessed sugar-sweetened soda consumption using a validated food frequency questionnaire (FFQ) at baseline. Diabetes was self-reported. We used Cox proportional hazard regression models to estimate the association between quintiles of sugar-sweetend soda and diabetes. We also estimated the associaiton by increasing one serving per day (355 mL) of sugar-sweetened soda. We conducted prespecified subgroup analysis by potential effect modifiers, namely markers of energy balance of early life factors, family history of diabetes, and Amerindian admixture.

RESULTS

During a median follow-up of 2.16 y (IQR 0.75-4.50) we identified 3,155 incident cases of diabetes. The median consumption of sugar-sweetened soda was 1.17 servings per day (IQR 0.47- 4.00). In multivariable analyses, comparing extreme quintiles showed that higher sugar-sweetened soda consumption was associated with diabetes incidence (HR = 1.32; 95% CI: 1.17, 1.49), and each additional serving per day of sugar-sweetened soda was associated with an increase of 27% in diabetes incidence (HR = 1.27; 95% CI: 1.16, 1.38). The soda-diabetes association was stronger among women who experienced intrauterine and childhood over-nutrition (high birth weight, no short stature, higher adiposity in premenarche, and higher adiposity at age 18-20 y old).

CONCLUSION

Sugar-sweetened soda consumption is associated with an increased risk of diabetes among Mexican women in a magnitude similar to that reported in other populations. The stronger association among individuals with markers of early life over-nutrition reinforce the need for early life interventions.

Comparative effectiveness of a bilayered living cellular construct and an acellular fetal bovine collagen dressing in the treatment of venous leg ulcers

Aim: To compare the effectiveness of bilayered living cellular construct (BLCC) and an acellular fetal bovine collagen dressing (FBCD) for the treatment of venous leg ulcers. Methods: Data from WoundExpert^® (Net Health, PA, USA) was used to analyze 1021 refractory venous leg ulcers treated at 177 facilities. Results: Kaplan-Meier analyses showed that BLCC (893 wounds) was superior to FBCD (128 wounds), p = 0.01 for: wound closure by weeks 12 (31 vs 25%), 24 (55 vs 43%) and 36 (68 vs 53%); reduction in time to wound closure of 37%, (19 vs 30 weeks); and improvement in the probability of healing by 45%. Conclusion: BLC versus FBCD showed significant differences in both time to and frequency of healing suggesting that BLCC may provide significant cost savings compared with FBCD.

Comparative Effectiveness of a Bioengineered Living Cellular Construct and Cryopreserved Cadaveric Skin Allograft for the Treatment of Venous Leg Ulcers in a Real-World Setting

Objective: A variety of advanced biological therapies are available for the treatment of chronic wounds such as venous leg ulcers (VLUs), but real-world comparative effectiveness data that can help guide decisions around treatments are currently lacking. Approach: This analysis was designed to compare the effectiveness of a bioengineered living cellular construct (BLCC) to a cryopreserved cadaveric skin allograft (CCSA) for the treatment of VLUs. Treatment records were collected from a large wound care-specific electronic medical record database on 717 patients (799 VLUs) receiving treatment at 177 wound care centers. Ulcers ≥28 days duration, between ≥1 and < 40 cm2 that closed ≤40% within the 28 days before treatment were included. Results: Patient baseline demographics and wound characteristics were comparable between groups. The median time to wound closure was 52% faster with BLCC compared with CCSA (15 weeks vs. 31 weeks). In addition, the proportion of wounds healed were significantly higher for BLCC by 12 weeks (42% vs. 24%) and 24 weeks (65% vs. 41%) (p = 0.0002). Treatment with BLCC increased the probability of healing by 97% compared with CCSA (hazard ratio = 1.97 [95% confidence interval 1.39-2.79], p = 0.0002). Innovation: This is the first real-world comparative effectiveness analysis to evaluate BLCC and CCSA for the treatment of VLUs. Conclusion: Treatment with a bioengineered cellular technology significantly improved the incidence and speed of wound closure compared with a CCSA.

Making fair comparisons in pregnancy medication safety studies: An overview of advanced methods for confounding control

Understanding the safety of medication use during pregnancy relies on observational studies: However, confounding in observational studies poses a threat to the validity of estimates obtained from observational data. Newer methods, such as marginal structural models and propensity calibration, have emerged to deal with complex confounding problems, but these methods have seen limited uptake in the pregnancy medication literature. In this article, we provide an overview of newer advanced methods for confounding control and show how these methods are relevant for pregnancy medication safety studies.

Handling missing data in propensity score estimation in comparative effectiveness evaluations: a systematic review

Aim

Even though systematic reviews have examined how aspects of propensity score methods are used, none has reviewed how the challenge of missing data is addressed with these methods. This review there-fore describes how missing data are addressed with propensity score methods in observational comparative effectiveness studies.

Methods

Published articles on observational comparative effectiveness studies were extracted from MEDLINE and EMBASE databases.

Results

Our search yielded 167 eligible articles. Majority of these studies (114; 68%) conducted complete case analysis with only 53 of them stating this in the methods. Only 16 articles reported use of multiple imputation.

Conclusion

Few researchers use correct methods for handling missing data or reported missing data methodology which may lead to reporting biased findings.

Risk of Ischemic Stroke Associated With Calcium Supplements With or Without Vitamin D: A Nested Case-Control Study

BACKGROUND

There is controversy surrounding the risk of ischemic stroke associated with the use of calcium supplements either in monotherapy or in combination with vitamin D.

METHODS AND RESULTS

A nested case-control study was performed with patients aged 40 to 89 years old, among whom a total of 2690 patients had a first episode of nonfatal ischemic stroke and for which 19 538 controls were randomly selected from the source population and frequency-matched with cases for age, sex, and calendar year. Logistic regression provided the odds ratios while adjusting for confounding factors. A sensitivity analysis was performed by restricting to patients who were new users of calcium supplements as either monotherapy or with vitamin D. Calcium supplementation with vitamin D was not associated with an increased risk of ischemic stroke (odds ratio 0.85; 95% confidence interval, 0.67-1.08) in the population as a whole or under any of the conditions examined (dose, duration, background cardiovascular risk, sex, or age). Calcium supplement monotherapy was not associated with an increased risk in the population as a whole (odds ratio 1.18; 95% confidence interval, 0.86-1.61), although a significant increased risk at high doses (≥1000 mg/day: odds ratio 2.09; 95% confidence interval, 1.25-3.49; <1000 mg: odds ratio 0.76; 95% confidence interval, 0.45-1.26) compared with nonuse was observed. The sensitivity analysis did not affect the inferences, with similar results observed among new users as to the overall study population.

CONCLUSIONS

This study suggests that calcium supplements given as monotherapy at high doses may increase the risk of ischemic stroke, whereas their combination with vitamin D seems to offset this hazard.

Incorporating linked healthcare claims to improve confounding control in a study of in-hospital medication use

INTRODUCTION

The Premier Perspective hospital billing database provides a promising data source for studies of inpatient medication use. However, in-hospital recording of confounders is limited, and incorporating linked healthcare claims data available for a subset of the cohort may improve confounding control. We investigated methods capable of adjusting for confounders measured in a subset, including complete case analysis, multiple imputation of missing data, and propensity score (PS) calibration.

METHODS

Methods were implemented in an example study of adults in Premier undergoing percutaneous coronary intervention (PCI) in 2004-2008 and exposed to either bivalirudin or heparin. In a subset of patients enrolled in UnitedHealth for at least 90 days before hospitalization, additional confounders were assessed from healthcare claims, including comorbidities, prior medication use, and service use intensity. Diagnostics for each method were evaluated, and methods were compared with respect to the estimates and confidence intervals of treatment effects on repeat PCI, bleeding, and in-hospital death.

RESULTS

Of 210,268 patients in the hospital-based cohort, 3240 (1.5 %) had linked healthcare claims. This subset was younger and healthier than the overall study population. The linked subset was too small for complete case evaluation of two of the three outcomes of interest. Multiple imputation and PS calibration did not meaningfully impact treatment effect estimates and associated confidence intervals.

CONCLUSIONS

Despite more than 98 % missingness on 24 variables, PS calibration and multiple imputation incorporated confounders from healthcare claims without major increases in estimate uncertainty. Additional research is needed to determine the relative bias of these methods.

Reverse Engineering and Evaluation of Prediction Models for Progression to Type 2 Diabetes: An Application of Machine Learning Using Electronic Health Records

BACKGROUND

Application of novel machine learning approaches to electronic health record (EHR) data could provide valuable insights into disease processes. We utilized this approach to build predictive models for progression to prediabetes and type 2 diabetes (T2D).

METHODS

Using a novel analytical platform (Reverse Engineering and Forward Simulation [REFS]), we built prediction model ensembles for progression to prediabetes or T2D from an aggregated EHR data sample. REFS relies on a Bayesian scoring algorithm to explore a wide model space, and outputs a distribution of risk estimates from an ensemble of prediction models. We retrospectively followed 24 331 adults for transitions to prediabetes or T2D, 2007-2012. Accuracy of prediction models was assessed using an area under the curve (AUC) statistic, and validated in an independent data set.

RESULTS

Our primary ensemble of models accurately predicted progression to T2D (AUC = 0.76), and was validated out of sample (AUC = 0.78). Models of progression to T2D consisted primarily of established risk factors (blood glucose, blood pressure, triglycerides, hypertension, lipid disorders, socioeconomic factors), whereas models of progression to prediabetes included novel factors (high-density lipoprotein, alanine aminotransferase, C-reactive protein, body temperature; AUC = 0.70).

CONCLUSIONS

We constructed accurate prediction models from EHR data using a hypothesis-free machine learning approach. Identification of established risk factors for T2D serves as proof of concept for this analytical approach, while novel factors selected by REFS represent emerging areas of T2D research. This methodology has potentially valuable downstream applications to personalized medicine and clinical research.

Misclassification in administrative claims data: quantifying the impact on treatment effect estimates

Misclassification is present in nearly every epidemiologic study, yet is rarely quantified in analysis in favor of a focus on random error. In this review, we discuss past and present wisdom on misclassification and what measures should be taken to quantify this influential bias, with a focus on bias in pharmacoepidemiologic studies. To date, pharmacoepidemiology primarily utilizes data obtained from administrative claims, a rich source of prescription data but susceptible to bias from unobservable factors including medication sample use, medications filled but not taken, health conditions that are not reported in the administrative billing data, and inadequate capture of confounders. Due to the increasing focus on comparative effectiveness research, we provide a discussion of misclassification in the context of an active comparator, including a demonstration of treatment effects biased away from the null in the presence of nondifferential misclassification. Finally, we highlight recently developed methods to quantify bias and offer these methods as potential options for strengthening the validity and quantifying uncertainty of results obtained from pharmacoepidemiologic research.

Risk of ischemic stroke associated with non-steroidal anti-inflammatory drugs and paracetamol: a population-based case-control study

OBJECTIVE

To assess the risk of non-fatal ischemic stroke associated with non-steroidal anti-inflammatory drugs (NSAIDs) and paracetamol. The effects of dose, duration of treatment, background cardiovascular (CV) risk and use of concomitant aspirin were studied.

METHODS

We performed a population-based case-control study. Patients were considered exposed if they were on treatment within a 30-day window before the index date. We estimated adjusted odds ratios (ORs) and their 95% confidence intervals (CIs) using logistic regression.

RESULTS

Two thousand eight hundred and eighty-eight cases and 20 000 controls were included. No increased risk was observed with traditional NSAIDs as a group (OR = 1.03; 95% CI, 0.90-1.19), but results varied across individual agents and conditions of use. An increased risk was found with diclofenac (OR = 1.53; 95% CI, 1.19-1.97), in particular when used at high doses (OR = 1.62; 1.06-2.46), over long-term periods (> 365 days; OR = 2.39; 1.52-3.76) and in patients with a high background CV risk (OR = 1.78; 1.23-2.58), as well as with aceclofenac when used at high doses (OR = 1.67; 1.05-2.67), in long-term treatments (OR = 2.00; 1.14-3.53) and in patients with CV risk factors (OR = 2.33; 1.40-3.87). No association was found with ibuprofen (OR = 0.94; 0.76-1.17) or naproxen (OR = 0.68; 0.36-1.29). The concomitant use of aspirin did not show a significant effect modification. Paracetamol did not increase the risk overall (OR = 0.97; 0.85-1.10) or in patients at high CV risk (OR = 0.94; 0.78-1.14).

CONCLUSIONS

Diclofenac and aceclofenac increase the risk of ischemic stroke while ibuprofen and naproxen do not. Dose, duration and baseline CV risk, but not aspirin use, appear to modulate the risk. Paracetamol does not increase the risk, even in patients with a high background CV risk.

Misclassification in administrative claims data: quantifying the impact on treatment effect estimates

Misclassification is present in nearly every epidemiologic study, yet is rarely quantified in analysis in favor of a focus on random error. In this review, we discuss past and present wisdom on misclassification and what measures should be taken to quantify this influential bias, with a focus on bias in pharmacoepidemiologic studies. To date, pharmacoepidemiology primarily utilizes data obtained from administrative claims, a rich source of prescription data but susceptible to bias from unobservable factors including medication sample use, medications filled but not taken, health conditions that are not reported in the administrative billing data, and inadequate capture of confounders. Due to the increasing focus on comparative effectiveness research, we provide a discussion of misclassification in the context of an active comparator, including a demonstration of treatment effects biased away from the null in the presence of nondifferential misclassification. Finally, we highlight recently developed methods to quantify bias and offer these methods as potential options for strengthening the validity and quantifying uncertainty of results obtained from pharmacoepidemiologic research.

Comparative effectiveness of a bilayered living cellular construct and a porcine collagen wound dressing in the treatment of venous leg ulcers

Using data from a national wound-specific electronic medical record (WoundExpert, Net Health, Pittsburgh, PA), we compared the effectiveness of a bilayered living cellular construct (BLCC) and an acellular porcine small intestine submucosa collagen dressing (SIS) for the treatment of venous leg ulcer. Data from 1,489 patients with 1,801 refractory venous leg ulcers (as defined by failure to have >40% reduction in size in the 4 weeks prior to treatment) with surface areas between 1 and 150 cm(2) in size, treated between July 2009 and July 2012 at 158 wound care facilities across the US were analyzed. Patient baseline demographics and wound characteristics were comparable between groups. Kaplan-Meier-derived estimates of wound closure for BLCC (1,451 wounds) was significantly greater (p = 0.01, log-rank test) by weeks 12 (31% vs. 26%), 24 (50% vs. 41%), and 36 (61% vs. 46%), respectively, compared with SIS (350 wounds). BLCC treatment reduced the median time to wound closure by 44%, achieving healing 19 weeks sooner (24 vs. 43 weeks, p = 0.01, log-rank test). Treatment with BLCC increased the probability of healing by 29% compared with porcine SIS dressing (hazard ratio = 1.29 [95% confidence interval 1.06, 1.56], p = 0.01).

Risk of nonfatal acute myocardial infarction associated with non-steroidal antiinflammatory drugs, non-narcotic analgesics and other drugs used in osteoarthritis: a nested case-control study

PURPOSE

The purpose of this study is to estimate the risk of nonfatal acute myocardial infarction (AMI) associated with traditional NSAIDs (tNSAIDs), non-narcotic analgesics (paracetamol and metamizole), and symptomatic slow-acting drugs in osteoarthritis (SYSADOAs) overall and in different subgroups of patients.

METHODS

We performed a nested case-control study using a Primary Care Database (Base de datos para la Investigación Farmacoepidemiológica en Atención Primaria), over the study period, 2001-2007. We included patients aged 40-90 years, with nonfatal AMI and randomly selected controls matched for age, sex and calendar year. Exposure to drugs was assessed within a 30-day window before the index date.

RESULTS

We did not find an association with nonfatal AMI in patients at low-intermediate background cardiovascular risk (odds ratio = 0.92; 95% confidence interval: 0.76-1.12), whereas there was a moderate significant association among those at high risk (1.28; 1.06-1.54) or when tNSAIDs were used for longer than 365 days (1.43; 1.12-1.82). The greatest risk occurred when these two conditions were combined (1.80; 1.26-2.58). The risk varied across individual tNSAIDs, with ibuprofen (0.95; 0.78-1.16) in the lower and aceclofenac (1.59; 1.15-2.19) in the upper part of the range. Low-dose aspirin did not modify the risk profile showed by any of the individual tNSAIDs examined. Paracetamol (0.84; 0.74-0.95), metamizole (1.06; 0.87-1.29) and SYSADOAs (0.68; 0.47-0.99) were not associated with an increased risk overall or in any subgroup of patients.

CONCLUSIONS

The risk of nonfatal AMI varied with individual tNSAIDs, duration of treatment and background cardiovascular risk. Paracetamol, metamizole and SYSADOAs did not increase the risk in any of the conditions examined.