Adjustment for continuous confounders: an example of how to prevent residual confounding

Food biodiversity and gastrointestinal cancer risk in nine European countries: Analysis within a prospective cohort study

BACKGROUND

Food biodiversity in human diets has potential co-benefits for both public health and sustainable food systems. However, current evidence on the potential relationship between food biodiversity and cancer risk, and particularly gastrointestinal cancers typically related to diet, remains limited. This study evaluated how dietary species richness (DSR) was associated with gastrointestinal cancer risk in a pan-European population.

METHODS

Associations between DSR and subsequent gastrointestinal cancer risk were examined among 450,111 adults enrolled in the European Prospective Investigation into Cancer and Nutrition cohort (EPIC, initiated in 1992), free of cancer at baseline. Usual dietary intakes were assessed at recruitment with country-specific dietary questionnaires. DSR of an individual's yearly diet was calculated based on the absolute number of unique biological species in each food and drink item. Associations between DSR and cancer risk were assessed by multivariable Cox proportional hazards regression models.

FINDINGS

During a median follow-up time of 14.1 years (SD=3.9), 10,705 participants were diagnosed with gastrointestinal cancer. Hazard ratios (HRs) and 95 % confidence intervals (CIs) comparing overall gastrointestinal cancer risk in the highest versus lowest quintiles of DSR indicated inverse associations in multivariable-adjusted models [HR (95 % CI): 0.77 (0.69-0.87); P-value < 0·0001] (Table 2). Specifically, inverse associations were observed between DSR and oesophageal squamous cell carcinoma, proximal colon, colorectal, and liver cancer risk (p-trend<0.05 for all cancer types).

INTERPRETATION

Greater food biodiversity in the diet may lower the risk of certain gastrointestinal cancers. Further research is needed to replicate these novel findings and to understand potential mechanisms.

Association between traumatic events with suicidality among adolescents: A large-scale cross-sectional study of 260,423 participants

Traumatic events have significant negative impacts throughout one's life. We aimed to comprehensively examine the early associations between traumatic events and suicidality among adolescents. In a cross-sectional sample of 260,423 adolescents in Deyang, China in September 2021, we assessed individual traumatic events, cumulative types, and patterns, alongside suicide risk scores and ideation, attempts, or plans. Linear and Poisson regression models adjusted for demographic confounders evaluated the association. Robust associations existed between interpersonal violence-related traumatic events and higher suicidality, with physical abuse demonstrating the strongest correlation. Moreover, suicide risk scores displayed a clear trend, indicating a progressively stronger association with suicidality as cumulative traumatic event types increased. Four distinct traumatic patterns emerged, including low traumas, high physical abuse, high death/serious injuries of a loved one, and multiple traumas, with the latter showing the strongest association with suicidality. Notably, the stratified analysis showed these associations were more pronounced in females, urban residents, only children, left-behind children, and those aged 13-15, while weaker in participants from families with intact parental relationships and middle socioeconomic status. Understanding the role of demographic factors and traumatic patterns in identifying at-risk youth can enable early detection and targeted interventions for suicide-related concerns.

Elevated homocysteine levels may moderate and mediate the association between HIV and cognitive impairment among middle-aged and older adults in an underserved population in Baltimore, Maryland

Background: In the antiretroviral therapy (ART) era, HIV-associated neurocognitive disorders (HAND) remain a considerable challenge for people with HIV, yet not all such disorders can be attributed to HIV alone. This study aimed to: (1) identify factors influencing neurocognitive impairment (NCI) utilizing the NIH Toolbox Cognition Battery (NIHTB-CB) as per the revised research criteria for HAND; (2) ascertain the moderating role of high homocysteine levels in the association between NCI and HIV; and (3) assess the mediating effect of elevated homocysteine levels on this association.Methods: We analyzed data from 788 adults (≥45 years) participating in a study on HIV-related comorbidities in underserved Baltimore communities, using NIHTB-CB to gauge neurocognitive performance. Special attention was given to results from the Dimensional Change Card Sort (DCCS) test within the executive function domain during causal mediation analysis.Results: Overall, HIV was not associated with NCI presence. However, HIV was associated with NCI among individuals with homocysteine >14 μmol/L. Furthermore, HIV was both directly and indirectly associated with NCI in DCCS test scores. Notably, the mediating role of elevated homocysteine in DCCS scores was only observable among individuals who had never used cocaine or had used it for ≤ 10 years, suggesting that extended cocaine use may have a substantial influence on cognitive performance.Conclusions: The findings from this study suggest elevated homocysteine levels may moderate and mediate the association between HIV and neurocognitive impairment.

Enzymatic debridement with bromelain and development of bacteremia in burn injuries: A retrospective cohort study

BACKGROUND

Debridement is crucial for effective wound management in patients with severe burn injuries, and bromelain, a proteolytic enzyme from pineapple stems, has emerged as a promising alternative for surgery. However, potential links of bromelain use to fever and sepsis have raised some concerns. Given the uncertainty as to whether this was caused by infection or other inflammatory sources, we aimed to investigate if the use of topical bromelain was associated with bacteremia.

METHODS

This single-centre retrospective cohort study included critically ill adult patients with severe burn injuries hospitalised at the Burn Center of the University Hospital Zurich between January 2017 and December 2021. Data were collected from two in-hospital electronic medical records databases. Our primary outcome, the association between topical bromelain treatment and the development of bacteremia, was investigated using a competing risk regression model, taking into account the competing risk of death. As a secondary outcome, the relationship between bromelain treatment and overall ICU mortality was examined using a Cox proportional hazards model.

RESULTS

The study included 269 patients with a median age of 50 years and median burnt total body surface area of 19%. A first bacteremia occurred in 61 patients (23%) after a median time of 6 days. Bromelain treatment was given to 83 (31%) of patients, with 22 (27%) of these developing bacteremia. In the fully adjusted competing risk regression model, no evidence for an association between bromelain treatment and bacteremia was found (SHR 0.79, 95%CI 0.42-1.48, p = 0.47). During hospital stay, 40 (15%) of patients died. There was no significant difference in mortality between patients treated with bromelain and those who were not (HR 0.55, 95%CI 0.26-1.20, p = 0.14). Among the five multidrug-resistant (MDR) pathogens identified, three were found in patients with bromelain treatment.

CONCLUSION

Our study did not confirm an association between topical bromelain and bacteremia in patients with severe burn injuries. This finding can inform evidence-based practices by addressing concerns about potential risks of bromelain use, contributing to the development of more effective and safe burn wound management strategies.

Incident cancers attributable to using opium and smoking cigarettes in the Golestan cohort study

Background

Opium consumption has recently been identified as a carcinogen, but the impact of opium use on cancer burden is unknown. We aimed to evaluate the fraction of cancers that could be attributed to opium use alone and in combination with cigarette smoking in a region where opium is widely used.

Methods

50,045 Iranian adults were recruited to this prospective cohort study between 2004 and 2008 and were followed through January 2022. We assessed the association between using opium and/or cigarette smoking and various cancers using proportional hazards regression models. We then calculated population attributable fractions (PAFs) for all cancers and for groups of cancers causally linked to opium and cigarette smoking.

Findings

Of the total participants, 8% only used opium, 8.3% only smoked cigarettes, and 9% used both substances. During a median 14 years of follow-up, 2195 individuals were diagnosed with cancer, including 215 opium-related cancers (lung, larynx, and bladder) and 1609 tobacco-related cancers (20 types). Opium use alone was estimated to cause 35% (95% CI: 26%-45%) of opium-related cancers, while smoking cigarettes alone was estimated to cause 9% (6%-12%) of tobacco-related cancers in this population. Using opium and/or cigarettes was estimated to cause 13% (9%-16%) of all cancers, 58% (49%-66%) of opium-related cancers, and 15% (11%-18%) of tobacco-related cancers. Moreover, joint exposure to opium and cigarettes had the greatest impact on cancers of the larynx, pharynx, lung, and bladder, with PAFs ranging from 50% to 77%.

Interpretation

Using opium and smoking cigarettes account for a large proportion of cancers in this population. To reduce the cancer burden, prevention policies should aim to decrease the use of both substances through public awareness campaigns and interventional efforts.

Funding

The Golestan Cohort Study work was funded by the Tehran University of Medical Sciences, Cancer Research UK, U.S. National Cancer Institute, International Agency for Research on Cancer. The presented analysis was supported by the International HundredK+ Cohorts Consortium (IHCC).

Association of caesarean delivery with offspring health outcomes in full-cohort versus sibling-comparison studies: a comparative meta-analysis and simulation study

BACKGROUND

Full-cohort and sibling-comparison designs have yielded inconsistent results about the impacts of caesarean delivery on offspring health outcomes, with the effect estimates from the latter being more likely directed towards the null value. We hypothesized that the seemingly conservative results obtained from the sibling-comparison design might be attributed to inadequate adjustment for non-shared confounders between siblings, particularly maternal age at delivery.

METHODS

A systematic review and meta-analysis was first conducted. PubMed, Embase, and the Web of Science were searched from database inception to April 6, 2022. Included studies (1) examined the association of caesarean delivery, whether elective or emergency, with offspring health outcomes; (2) simultaneously conducted full-cohort and sibling-comparison analyses; and (3) reported adjusted effect estimates with 95% confidence intervals (95% CIs). No language restrictions were applied. Data were extracted by 2 reviewers independently. Three-level meta-analytic models were used to calculate the pooled odds ratios (ORs) and 95% CIs for caesarean versus vaginal delivery on multiple offspring health outcomes separately for full-cohort and sibling-comparison designs. Subgroup analyses were performed based on the method of adjustment for maternal age at delivery. A simulation study was then conducted. The simulated datasets were generated with some key parameters derived from the meta-analysis.

RESULTS

Eighteen studies involving 21,854,828 individuals were included. The outcomes assessed included mental and behavioral disorders; endocrine, nutritional and metabolic diseases; asthma; cardiorespiratory fitness; and multiple sclerosis. The overall pooled OR for estimates from the full-cohort design was 1.14 (95% CI: 1.11 to 1.17), higher than that for estimates from the sibling-comparison design (OR = 1.08; 95% CI: 1.02 to 1.14). Stratified analyses showed that estimates from the sibling-comparison design varied considerably across studies using different methods to adjust for maternal age at delivery in multivariate analyses, while those from the full-cohort design were rather stable: in studies that did not adjust maternal age at delivery, the pooled OR of full-cohort vs. sibling-comparison design was 1.10 (95% CI: 0.99 to 1.22) vs. 1.06 (95% CI: 0.85 to 1.31), in studies adjusting it as a categorical variable, 1.15 (95% CI: 1.11 to 1.19) vs. 1.07 (95% CI: 1.00 to 1.15), and in studies adjusting it as a continuous variable, 1.12 (95% CI: 1.05 to 1.19) vs. 1.12 (95% CI: 0.98 to 1.29). The severe underestimation bias related to the inadequate adjustment of maternal age at delivery in sibling-comparison analyses was fully replicated in the simulation study.

CONCLUSIONS

Sibling-comparison analyses may underestimate the association of caesarean delivery with multiple offspring health outcomes due to inadequate adjustment of non-shared confounders, such as maternal age at delivery. Thus, we should be cautious when interpreting the seemingly conservative results of sibling-comparison analyses in delivery-related studies.

Challenges in synthesis of real-world vaccine effects on meningococcal serogroup B disease for 4CMenB vaccine post-licensure effectiveness studies: A systematic review

BACKGROUND

Real-world studies on vaccine effects are diverse in terms of objectives, study setting and design, data type and scope, and analysis methods. In this review, we describe and discuss four-component meningococcal serogroup B vaccine (4CMenB vaccine, Bexsero) real-world studies with the aim of synthesizing their findings with application of standard methods.

METHODS

We performed a systematic literature review of all real-world studies on 4CMenB vaccine effects on meningococcal serogroup B disease, with no restriction for population age, vaccination schedule and/or type of vaccine effect evaluated (vaccine effectiveness [VE] and vaccine impact [VI] outcomes) published since its licensure in 2013 (from January 2014 until July 2021) in PubMed, Cochrane and the grey literature. We then aimed to synthesize the findings of the identified studies through application of standard synthesis methods.

RESULTS

According to reported criteria we retrieved five studies presenting estimates on 4CMenB vaccine effectiveness and impact. These studies showed great diversity in population, vaccination schedule and analysis methods mainly due to diversity in vaccine strategies and recommendations in the study settings. Directed by this diversity, no quantitative pooling methods to synthesize findings could be applied; instead we descriptively assessed study methods. We report VE estimates ranging from 59% to 94% and VI estimates ranging from 31% to 75%, representing diverse age groups, vaccination schedules and analysis methods.

CONCLUSION

Both vaccine outcomes showed real-life effectiveness of 4CMenB vaccine despite differences in study methodologies and vaccination strategies. Based on appraisal of study methods, we highlighted the need for an adapted tool which facilitates synthesis of heterogenic real-world vaccine studies when quantitative pooling methods are not applicable.

Cocaine Use May Moderate the Associations of HIV and Female Sex with Neurocognitive Impairment in a Predominantly African American Population Disproportionately Impacted by HIV and Substance Use

HIV-associated neurocognitive disorders (HAND) remain a major challenge for people with HIV in the antiretroviral therapy era. Cocaine use may trigger/exacerbate HAND among African American (AA) adults, especially women. Between 2018 and 2019, 922 adults, predominantly AAs, with/without HIV and with/without cocaine use in Baltimore, Maryland, were enrolled in a study investigating the association of HIV and cocaine use with neurocognitive impairment (NCI). Neurocognitive performance was assessed with the NIH Toolbox Cognition Battery (NIHTB-CB). NCI was considered to be present if the fully adjusted standard score for at least two cognitive domains was 1.0 standard deviation below the mean. Although the overall analysis showed HIV and female sex were associated with NCI, the associations were dependent on cocaine use. Neither HIV [adj prevalence ratio (PR): 1.12, confidence interval (95% CI): 0.77-1.64] nor female sex (adj PR: 1.07, 95% CI: 0.71-1.61) was associated with NCI among cocaine nonusers, while both HIV (adj PR: 1.39, 95% CI: 1.06-1.81) and female sex (adj PR: 1.53, 95% CI: 1.18-1.98) were associated with NCI in cocaine users. HIV was associated with two NIHTB-CB measures overall. In addition, HIV was associated with a lower dimensional change card sort score (an executive function measure) in cocaine users and not in nonusers. Cognitive performance was poorer in female than in male cocaine users. The adverse effect of HIV on cognitive performance predominantly affected cocaine users. However, cocaine use may moderate the impact of HIV and female sex on cognitive performance, highlighting the importance of reducing cocaine use in NCI prevention among the AA population.

Effects of influential points and sample size on the selection and replicability of multivariable fractional polynomial models

BACKGROUND

The multivariable fractional polynomial (MFP) approach combines variable selection using backward elimination with a function selection procedure (FSP) for fractional polynomial (FP) functions. It is a relatively simple approach which can be easily understood without advanced training in statistical modeling. For continuous variables, a closed test procedure is used to decide between no effect, linear, FP1, or FP2 functions. Influential points (IPs) and small sample sizes can both have a strong impact on a selected function and MFP model.

METHODS

We used simulated data with six continuous and four categorical predictors to illustrate approaches which can help to identify IPs with an influence on function selection and the MFP model. Approaches use leave-one or two-out and two related techniques for a multivariable assessment. In eight subsamples, we also investigated the effects of sample size and model replicability, the latter by using three non-overlapping subsamples with the same sample size. For better illustration, a structured profile was used to provide an overview of all analyses conducted.

RESULTS

The results showed that one or more IPs can drive the functions and models selected. In addition, with a small sample size, MFP was not able to detect some non-linear functions and the selected model differed substantially from the true underlying model. However, when the sample size was relatively large and regression diagnostics were carefully conducted, MFP selected functions or models that were similar to the underlying true model.

CONCLUSIONS

For smaller sample size, IPs and low power are important reasons that the MFP approach may not be able to identify underlying functional relationships for continuous variables and selected models might differ substantially from the true model. However, for larger sample sizes, a carefully conducted MFP analysis is often a suitable way to select a multivariable regression model which includes continuous variables. In such a case, MFP can be the preferred approach to derive a multivariable descriptive model.

Omicron Subvariants: Clinical, Laboratory, and Cell Culture Characterization

BACKGROUND

The variant of concern Omicron has become the sole circulating severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) variant for the past several months. Omicron subvariants BA.1, BA.2, BA.3, BA.4, and BA.5 evolved over the time, with BA.1 causing the largest wave of infections globally in December 2021-January 2022. This study compared the clinical outcomes in patients infected with different Omicron subvariants and the relative viral loads and recovery of infectious virus from upper respiratory specimens.

METHODS

SARS-CoV-2-positive remnant clinical specimens, diagnosed at the Johns Hopkins Microbiology Laboratory between December 2021 and July 2022, were used for whole-genome sequencing. The clinical outcomes of infections with Omicron subvariants were compared with infections with BA.1. Cycle threshold (Ct) values and the recovery of infectious virus on the VeroTMPRSS2 cell line from clinical specimens were compared.

RESULTS

BA.1 was associated with the largest increase in SARS-CoV-2 positivity rate and coronavirus disease 2019 (COVID-19)-related hospitalizations at the Johns Hopkins system. After a peak in January, cases decreased in the spring, but the emergence of BA.2.12.1 followed by BA.5 in May 2022 led to an increase in case positivity and admissions. BA.1 infections had a lower mean Ct value when compared with other Omicron subvariants. BA.5 samples had a greater likelihood of having infectious virus at Ct values <20.

CONCLUSIONS

Omicron subvariants continue to be associated with a relatively high rate of polymerase chain reaction (PCR) positivity and hospital admissions. The BA.5 infections are more while BA.2 infections are less likely to have infectious virus, suggesting potential differences in infectibility during the Omicron waves.

Confounding in observational studies evaluating the association between Alzheimer's disease and periodontal disease: A systematic review

Background

Studies investigating the association between periodontitis and Alzheimer's disease (AD) suggested indirect (periodontitis would increase the circulation of inflammation-inducible molecules) and direct (periodontopathogens might colonize brains affected by Alzheimer's disease) pathways. While there seems to be a positive relationship between periodontitis and AD, concerns have been raised about the role of confounding.

Aim

To systematically review the literature to assess confounding and their level of heterogeneity in the association between periodontitis and AD. Also, to examine data reporting and interpretation regarding confounding bias.

Methods

This review followed the PRISMA guidelines and was registered within PROSPERO. Electronic searches were performed in seven main databases and three others to capture the "grey literature". The PECO strategy was used to identify observational studies (cross-sectional, case-control, or cohort studies) assessing the association between periodontal disease and AD without restricting publication language and year. Critical appraisal was performed according to the Joanna Briggs Institute guidelines. Confounders were evaluated following a two-step approach.

Results

A total of 3255 studies were found, of which 18 (13 case-control, four cross-sectional, and one cohort) met the eligibility criteria. Participants with AD were 1399 (mean age 64 ± 9 to 84.8 ± 5.6 years), whereas those without AD were 1730 (mean age 62.6 ± 7.1 to 81.4 ± 4.6). Female patients composed most of the sample for both groups. The confounding variables "age" and "sex" were present in all studies. Four studies used the 2017 AAP/EFP periodontal classification. Most studies had a low risk of bias. Fifty percent of the articles did not consider confounding; variation in the adjustment approaches was observed. Additionally, 62% of the studies did not mention bias, and 40% did not discuss any limitations about confounders.

Conclusions

Given the study's limitations, caution must be taken to properly interpret the association between periodontitis and AD.Registration: CRD42022293884.

Misspecification of confounder-exposure and confounder-outcome associations leads to bias in effect estimates

BACKGROUND

Confounding is a common issue in epidemiological research. Commonly used confounder-adjustment methods include multivariable regression analysis and propensity score methods. Although it is common practice to assess the linearity assumption for the exposure-outcome effect, most researchers do not assess linearity of the relationship between the confounder and the exposure and between the confounder and the outcome before adjusting for the confounder in the analysis. Failing to take the true non-linear functional form of the confounder-exposure and confounder-outcome associations into account may result in an under- or overestimation of the true exposure effect. Therefore, this paper aims to demonstrate the importance of assessing the linearity assumption for confounder-exposure and confounder-outcome associations and the importance of correctly specifying these associations when the linearity assumption is violated.

METHODS

A Monte Carlo simulation study was used to assess and compare the performance of confounder-adjustment methods when the functional form of the confounder-exposure and confounder-outcome associations were misspecified (i.e., linearity was wrongly assumed) and correctly specified (i.e., linearity was rightly assumed) under multiple sample sizes. An empirical data example was used to illustrate that the misspecification of confounder-exposure and confounder-outcome associations leads to bias.

RESULTS

The simulation study illustrated that the exposure effect estimate will be biased when for propensity score (PS) methods the confounder-exposure association is misspecified. For methods in which the outcome is regressed on the confounder or the PS, the exposure effect estimate will be biased if the confounder-outcome association is misspecified. In the empirical data example, correct specification of the confounder-exposure and confounder-outcome associations resulted in smaller exposure effect estimates.

CONCLUSION

When attempting to remove bias by adjusting for confounding, misspecification of the confounder-exposure and confounder-outcome associations might actually introduce bias. It is therefore important that researchers not only assess the linearity of the exposure-outcome effect, but also of the confounder-exposure or confounder-outcome associations depending on the confounder-adjustment method used.

Association between primary or booster COVID-19 mRNA vaccination and Omicron lineage BA.1 SARS-CoV-2 infection in people with a prior SARS-CoV-2 infection: A test-negative case-control analysis

BACKGROUND

The benefit of primary and booster vaccination in people who experienced a prior Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) infection remains unclear. The objective of this study was to estimate the effectiveness of primary (two-dose series) and booster (third dose) mRNA vaccination against Omicron (lineage BA.1) infection among people with a prior documented infection.

METHODS AND FINDINGS

We conducted a test-negative case-control study of reverse transcription PCRs (RT-PCRs) analyzed with the TaqPath (Thermo Fisher Scientific) assay and recorded in the Yale New Haven Health system from November 1, 2021, to April 30, 2022. Overall, 11,307 cases (positive TaqPath analyzed RT-PCRs with S-gene target failure [SGTF]) and 130,041 controls (negative TaqPath analyzed RT-PCRs) were included (median age: cases: 35 years, controls: 39 years). Among cases and controls, 5.9% and 8.1% had a documented prior infection (positive SARS-CoV-2 test record ≥90 days prior to the included test), respectively. We estimated the effectiveness of primary and booster vaccination relative to SGTF-defined Omicron (lineage BA.1) variant infection using a logistic regression adjusted for date of test, age, sex, race/ethnicity, insurance, comorbidities, social venerability index, municipality, and healthcare utilization. The effectiveness of primary vaccination 14 to 149 days after the second dose was 41.0% (95% confidence interval (CI): 14.1% to 59.4%, p 0.006) and 27.1% (95% CI: 18.7% to 34.6%, p < 0.001) for people with and without a documented prior infection, respectively. The effectiveness of booster vaccination (≥14 days after booster dose) was 47.1% (95% CI: 22.4% to 63.9%, p 0.001) and 54.1% (95% CI: 49.2% to 58.4%, p < 0.001) in people with and without a documented prior infection, respectively. To test whether booster vaccination reduced the risk of infection beyond that of the primary series, we compared the odds of infection among boosted (≥14 days after booster dose) and booster-eligible people (≥150 days after second dose). The odds ratio (OR) comparing boosted and booster-eligible people with a documented prior infection was 0.79 (95% CI: 0.54 to 1.16, p 0.222), whereas the OR comparing boosted and booster-eligible people without a documented prior infection was 0.54 (95% CI: 0.49 to 0.59, p < 0.001). This study's limitations include the risk of residual confounding, the use of data from a single system, and the reliance on TaqPath analyzed RT-PCR results.

CONCLUSIONS

In this study, we observed that primary vaccination provided significant but limited protection against Omicron (lineage BA.1) infection among people with and without a documented prior infection. While booster vaccination was associated with additional protection against Omicron BA.1 infection in people without a documented prior infection, it was not found to be associated with additional protection among people with a documented prior infection. These findings support primary vaccination in people regardless of documented prior infection status but suggest that infection history may impact the relative benefit of booster doses.

Omicron Subvariants: Clinical, Laboratory, and Cell Culture Characterization

Background

The variant of concern, Omicron, has become the sole circulating SARS-CoV-2 variant for the past several months. Omicron subvariants BA.1, BA.2, BA.3, BA.4, and BA.5 evolved over the time, with BA.1 causing the largest wave of infections globally in December 2021- January 2022. In this study, we compare the clinical outcomes in patients infected with different Omicron subvariants and compare the relative viral loads, and recovery of infectious virus from upper respiratory specimens.

Methods

SARS-CoV-2 positive remnant clinical specimens, diagnosed at the Johns Hopkins Microbiology Laboratory between December 2021 and July 2022, were used for whole genome sequencing. The clinical outcomes of infections with Omicron subvariants were compared to infections with BA.1. Cycle threshold values (Ct) and the recovery of infectious virus on VeroTMPRSS2 cell line from clinical specimens were compared.

Results

The BA.1 was associated with the largest increase in SARS-CoV-2 positivity rate and COVID-19 related hospitalizations at the Johns Hopkins system. After a peak in January cases fell in the spring, but the emergence of BA.2.12.1 followed by BA.5 in May 2022 led to an increase in case positivity and admissions. BA.1 infections had a lower mean Ct when compared to other Omicron subvariants. BA.5 samples had a greater likelihood of having infectious virus at Ct values less than 20.

Conclusions

Omicron subvariants continue to associate with a relatively high positivity and admissions. The BA.5 infections are more while BA.2 infections are less likely to have infectious virus, suggesting potential differences in infectibility during the Omicron waves.

Funding

Centers for Disease Control and Prevention contract 75D30121C11061, NIH/NIAID Center of Excellence in Influenza Research and Surveillance contract HHS N2772201400007C, Johns Hopkins University, Maryland department of health, and The Modeling Infectious Diseases in Healthcare Network (MInD) under awards U01CK000589.

Machine learning nonresponse adjustment of patient-reported opioid consumption data to enable consumption-informed postoperative opioid prescribing guidelines

Background

Post-discharge opioid consumption is a crucial patient-reported outcome informing opioid prescribing guidelines, but its collection is resource-intensive and vulnerable to inaccuracy due to nonresponse bias.

Methods

We developed a post-discharge text message-to-web survey system for efficient collection of patient-reported pain outcomes. We prospectively recruited surgical patients at Beth Israel Deaconess Medical Center in Boston, Massachusetts from March 2019 through October 2020, sending an SMS link to a secure web survey to quantify opioids consumed after discharge from hospitalization. Patient factors extracted from the electronic health record were tested for nonresponse bias and observable confounding. Following targeted learning-based nonresponse adjustment, procedure-specific opioid consumption quantiles (medians and 75th percentiles) were estimated and compared to a previous telephone-based reference survey.

Results

6553 patients were included. Opioid consumption was measured in 44% of patients (2868), including 21% (1342) through survey response. Characteristics associated with inability to measure opioid consumption included age, tobacco use, and prescribed opioid dose. Among the 10 most common procedures, median consumption was only 36% of the median prescription size; 64% of prescribed opioids were not consumed. Among those procedures, nonresponse adjustment corrected the median opioid consumption by an average of 37% (IQR: 7, 65%) compared to unadjusted estimates, and corrected the 75th percentile by an average of 5% (IQR: 0, 12%). This brought median estimates for 5/10 procedures closer to telephone survey-based consumption estimates, and 75th percentile estimates for 2/10 procedures closer to telephone survey-based estimates.

Conclusions

SMS-recruited online surveying can generate reliable opioid consumption estimates after nonresponse adjustment using patient factors recorded in the electronic health record, protecting patients from the risk of inaccurate prescription guidelines.

Modeling non-linear relationships in epidemiological data: The application and interpretation of spline models

Objective

Traditional methods to deal with non-linearity in regression analysis often result in loss of information or compromised interpretability of the results. A recommended but underutilized method for modeling non-linear associations in regression models is spline functions. We explain spline functions in a non-mathematical way and illustrate the application and interpretation to an empirical data example.

Methods

Using data from the Amsterdam Growth and Health Longitudinal Study, we examined the non-linear relationship between the sum of four skinfolds and VO2max, which are measures of body fat and cardiorespiratory fitness, respectively. We compared traditional methods (i.e., quadratic regression and categorization) to spline methods [1- and 3-knot linear spline (LSP) models and a 3-knot restricted cubic spline (RCS) model] in terms of the interpretability of the results and their explained variance (r a d j 2 ).

Results

The spline models fitted the data better than the traditional methods. Increasing the number of knots in the LSP model increased the explained variance (from r a d j 2 = 0 . 578 for the 1-knot model to r a d j 2 = 0 . 582 for the 3-knot model). The RCS model fitted the data best (r a d j 2 = 0 . 591 ), but results in regression coefficients that are harder to interpret.

Conclusion

Spline functions should be considered more often as they are flexible and can be applied in commonly used regression analysis. RCS regression is generally recommended for prediction research (i.e., to obtain the predicted outcome for a specific exposure value), whereas LSP regression is recommended if one is interested in the effects in a population.

Is premature ovarian insufficiency associated with mortality? A three-decade follow-up cohort

OBJECTIVE

To evaluate the association between premature ovarian insufficiency (POI) and mortality.

MATERIALS AND METHODS

This was a secondary analysis of a long-term cohort of Chilean women who received preventive health care between 1990 and 1993. The exposure variable was POI and the outcome was death, and follow-up time was 30 years. Patient data were extracted from medical records. Data related to deaths were obtained from the records of the official government registry as of January 2021. Cox regression proportional hazard models were used to estimate crude and adjusted hazard ratios (HR) and 95 % confidence intervals (CI).

RESULTS

Data for a total of 1119 women were included in the analysis. Median age was 47 years (interquartile range: 44-52). The baseline prevalence of POI was 6.7 %. At the end of the follow-up, 34.7 % of women with POI had died, compared with 19.3 % of women without the condition (p < 0.001). A larger proportion of women with POI died from cardiovascular disease (12.0 % vs. 5.1 %; OR: 2.55, 95 % CI: 1.21-5.39) whereas there was no significant difference in cancer mortality (6.7 % vs. 7.7 %; OR: 0.86, 95 % CI: 0.34-2.19). In the adjusted Cox model, POI was among the main factors associated with mortality (hazard ratio [HR] 1.60, 95 % CI: 1.03-2.47), after diabetes (HR 2.51, 95 % CI: 1.40-4.51) and arterial hypertension (HR 1.75, 95 % CI: 1.29-2.37).

CONCLUSION

Although POI affects a small group of women, its association with mortality seems to be relevant; hence it is necessary to implement measures that reduce this risk.

Better COVID-19 Intensive Care Unit survival in females, independent of age, disease severity, comorbidities, and treatment

Although male Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) patients have higher Intensive Care Unit (ICU) admission rates and a worse disease course, a comprehensive analysis of female and male ICU survival and underlying factors such as comorbidities, risk factors, and/or anti-infection/inflammatory therapy administration is currently lacking. Therefore, we investigated the association between sex and ICU survival, adjusting for these and other variables. In this multicenter observational cohort study, all patients with SARS-CoV-2 pneumonia admitted to seven ICUs in one region across Belgium, The Netherlands, and Germany, and requiring vital organ support during the first pandemic wave were included. With a random intercept for a center, mixed-effects logistic regression was used to investigate the association between sex and ICU survival. Models were adjusted for age, Acute Physiology and Chronic Health Evaluation II (APACHE II) score, comorbidities, and anti-infection/inflammatory therapy. Interaction terms were added to investigate effect modifications by sex with country and sex with obesity. A total of 551 patients (29% were females) were included. Mean age was 65.4 ± 11.2 years. Females were more often obese and smoked less frequently than males (p-value 0.001 and 0.042, respectively). APACHE II scores of females and males were comparable. Overall, ICU mortality was 12% lower in females than males (27% vs 39% respectively, p-value < 0.01) with an odds ratio (OR) of 0.62 (95%CI 0.39-0.96, p-value 0.032) after adjustment for age and APACHE II score, 0.63 (95%CI 0.40-0.99, p-value 0.044) after additional adjustment for comorbidities, and 0.63 (95%CI 0.39-0.99, p-value 0.047) after adjustment for anti-infection/inflammatory therapy. No effect modifications by sex with country and sex with obesity were found (p-values for interaction > 0.23 and 0.84, respectively). ICU survival in female SARS-CoV-2 patients was higher than in male patients, independent of age, disease severity, smoking, obesity, comorbidities, anti-infection/inflammatory therapy, and country. Sex-specific biological mechanisms may play a role, emphasizing the need to address diversity, such as more sex-specific prediction, prognostic, and therapeutic approach strategies.

Cut points and contexts

In research, policy, and practice, continuous variables are often categorized. Statisticians have generally advised against categorization for many reasons, such as loss of information and precision as well as distortion of estimated statistics. Here, a different kind of problem with categorization is considered: the idea that, for a given continuous variable, there is a unique set of cut points that is the objectively correct or best categorization. It is shown that this is unlikely to be the case because categorized variables typically exist in webs of statistical relationships with other variables. The choice of cut points for a categorized variable can influence the values of many statistics relating that variable to others. This essay explores the substantive trade‐offs that can arise between different possible cut points to categorize a continuous variable, making it difficult to say that any particular categorization is objectively best. Limitations of different approaches to selecting cut points are discussed. Contextual trade‐offs may often be an argument against categorization. At the very least, such trade‐offs mean that research inferences, or decisions about policy or practice, that involve categorized variables should be framed and acted upon with flexibility and humility.

In practical settings, the choice of cut points for categorizing a continuous variable is likely to entail trade‐offs across multiple statistical relationships between the categorized variable and other variables. These trade‐offs mean that no single categorization is objectively best or correct.

Improving the investigative approach to polycythaemia vera: a critical assessment of current evidence and vision for the future

Polycythaemia vera is a challenging disease to study given its low prevalence and prolonged time-to-event for important clinical endpoints such as thrombosis, progression, and mortality. Although researchers in this space often rise to meet these challenges, there is considerable room for improvement in the analysis of retrospective data, the development of risk-stratification tools, and the design of randomised controlled trials. In this Viewpoint, we review the evidence behind the contemporary approach to risk stratification and treatment of polycythaemia vera. Frameworks for using data more efficiently, constructing more nuanced prognostic models, and overcoming challenges in clinical trial design are discussed.

Paradoxical association between atrial fibrillation/flutter and high cholesterol over age 75 years: The Kuakini Honolulu Heart Program and Honolulu-Asia Aging Study

INTRODUCTION

Several studies have indicated high cholesterol is paradoxically associated with low prevalence of atrial fibrillation/flutter (AF). However, the etiology is uncertain. One potential explanation might be the confounding effect of age exemplifying prevalence-incidence (Neyman's) bias. However, this bias has not often been discussed in depth in the literature. Therefore, we conducted a cross-sectional analysis to test the hypothesis that there is a paradoxical association between lipid profile and AF prevalence.

METHODS

This is a cross-sectional study design, using data from the Kuakini Honolulu Heart Program. Participants were 3741 Japanese-American men between 71 and 93 years old living in Hawaii. Serum total cholesterol (TC) level was measured and categorized into quartiles. AF was diagnosed by 12‑lead Electrocardiogram. We categorized age into quartiles (71-74, 75-77, 78-80 and 81+ years).

RESULTS

We observed opposite associations between AF and TC among different age groups. For participants age ≥75, higher TC levels were paradoxically associated with lower prevalence of AF after multivariable adjustment, i.e. the odds ratios of AF comparing the highest TC quartile with the lowest TC quartile for age 75-77, 78-80 and 81+ years were 0.17 (95% confidence interval [CI], 0.06-0.52), 0.28 (95% CI, 0.07-1.09) and 0.14 (95% CI, 0.03-0.62), respectively. Conversely, for those who were 71-74 years old, the odds ratio of AF was 2.09 (95% CI, 0.76-5.75) between the highest and the lowest TC quartiles.

CONCLUSIONS

There is a paradoxical association of TC with AF in Japanese-American men age ≥75, but not <75 years. The paradox might be explained by Neyman's bias.

Coronary heart disease and stroke in the Sami and non-Sami populations in rural Northern and Mid Norway-the SAMINOR Study

BACKGROUND

Previous studies have suggested that Sami have a similar risk of myocardial infarction and a higher risk of stroke compared with non-Sami living in the same geographical area.

DESIGN

Participants in the SAMINOR 1 Survey (2003-2004) aged 30 and 36-79 years were followed to the 31 December 2016 for observation of fatal or non-fatal events of acute myocardial infarction (AMI), coronary heart disease (CHD), ischaemic stroke (IS), stroke and a composite endpoint (fatal or non-fatal AMI or stroke).

AIM

Compare the risk of AMI, CHD, IS, stroke and the composite endpoint in Sami and non-Sami populations, and identify intermediate factors if ethnic differences in risks are observed.

METHODS

Cox regression models.

RESULTS

The sex-adjusted and age-adjusted risks of AMI (HR for Sami versus non-Sami 0.99, 95% CI: 0.83 to 1.17), CHD (HR 1.03, 95% CI: 0.93 to 1.15) and of the composite endpoint (HR 1.09, 95% CI: 0.95 to 1.24) were similar in Sami and non-Sami populations. Sami ethnicity was, however, associated with increased risk of IS (HR 1.36, 95% CI: 1.10 to 1.68) and stroke (HR 1.31, 95% CI: 1.08 to 1.58). Height explained more of the excess risk observed in Sami than conventional risk factors.

CONCLUSIONS

The risk of IS and stroke were higher in Sami and height was identified as an important intermediate factor as it explained a considerable proportion of the ethnic differences in IS and stroke. The risk of AMI, CHD and the composite endpoint was similar in Sami and non-Sami populations.

Impact of the Improper Adjustment for Age in Research on Age-Related Macular Degeneration: An Example Using Data from the Canadian Longitudinal Study on Aging

Purpose: Confounding is an important problem in observational research. Improper modeling of the confounder will lead to residual confounding that may distort results and impact inferences. An example of this will be presented from research on age-related macular degeneration and depression. Methods: A 3-year prospective cohort study was performed using data from the Canadian Longitudinal Study on Aging consisting of 30,097 individuals aged 45-85 years. Incident depression was assessed using the Center for Epidemiologic Studies Depression scale. Participants were asked if they had ever had a physician diagnosis of age-related macular degeneration (AMD). Multivariable Poisson regression was used. Age was modeled in four ways including as a linear term, as a 4-category variable, as a spline, and as a polynomial. Models were compared using the Akaike's Information Criteria (AIC) with lower scores indicating better performance. Results: The point estimates and inferences differed depending on how age was modeled. Age had a J-shape relationship with the incidence of depression. The model with the lowest AIC was when age was entered as a categorical variable. When age was modeled in this way, AMD was not significantly associated with the incidence of depression (relative risk (RR) = 1.21, 95% Confidence Interval (CI) 0.97, 1.53). By contrast, when age was modeled as a linear term, AMD was significantly associated with the incidence of depression (RR = 1.28, 95% CI 1.02, 1.61). Conclusions: Researchers should clearly report their adjustment strategies and should be cautious when modeling the relationship between age and depression in order to minimize residual confounding.

Adverse effects of Z-drugs for sleep disturbance in people living with dementia: a population-based cohort study

BACKGROUND

Sleep disturbance is common in dementia and often treated with Z-drugs (zopiclone, zaleplon, and zolpidem). While some observational studies suggest that Z-drugs are associated with adverse events such as falls and fracture risks in older people, this has not been studied in dementia.

METHODS

We used data from 27,090 patients diagnosed with dementia between January 2000 and March 2016 from the Clinical Practice Research Datalink linked to Hospital Episodes Statistics data in England. We compared adverse events for 3532 patients newly prescribed Z-drugs by time-varying dosage to (1) 1833 non-sedative-users with sleep disturbance; (2) 10,214 non-sedative-users with proximal GP consultation matched on age, sex, and antipsychotic use; and (3) 5172 patients newly prescribed benzodiazepines. We defined higher dose Z-drugs and benzodiazepines as prescriptions equivalent to ≥ 7.5 mg zopiclone or > 5 mg diazepam daily. Cox regression was used to estimate hazard ratios (HRs) for incident fracture, hip fracture, fall, mortality, acute bacterial infection, ischaemic stroke/transient ischaemic attack, and venous thromboembolism over a 2-year follow-up, adjusted for demographic- and health-related covariates.

RESULTS

The mean (SD) age of patients was 83 (7.7) years, and 16,802 (62%) were women. Of 3532 patients prescribed Z-drugs, 584 (17%) were initiated at higher doses. For patients prescribed higher dose Z-drugs relative to non-users with sleep disturbance, the HRs (95% confidence interval) for fractures, hip fractures, falls, and ischaemic stroke were 1.67 (1.13-2.46), 1.96 (1.16-3.31), 1.33 (1.06-1.66), and 1.88 (1.14-3.10), respectively. We observed similar associations when compared to non-sedative-users with proximal GP consultation. Minimal or inconsistent excess risks were observed at ≤ 3.75 mg zopiclone or equivalent daily, and for mortality, infection, and venous thromboembolism. We observed no differences in adverse events for Z-drugs compared to benzodiazepines, except lower mortality rates with Z-drugs (HR [95% confidence interval] of 0.73 [0.64-0.83]).

CONCLUSIONS

Higher dose Z-drug use in dementia is associated with increased fracture and stroke risks, similar or greater to that for higher dose benzodiazepines. Higher dose Z-drugs should be avoided, if possible, in people living with dementia, and non-pharmacological alternatives preferentially considered. Prescriptions for higher dose Z-drugs in dementia should be regularly reviewed.

TRIAL REGISTRATION

ENCePP e-register of studies, EUPAS18006.

Approaches to addressing missing values, measurement error, and confounding in epidemiologic studies

OBJECTIVES

Epidemiologic studies often suffer from incomplete data, measurement error (or misclassification), and confounding. Each of these can cause bias and imprecision in estimates of exposure-outcome relations. We describe and compare statistical approaches that aim to control all three sources of bias simultaneously.

STUDY DESIGN AND SETTING

We illustrate four statistical approaches that address all three sources of bias, namely, multiple imputation for missing data and measurement error, multiple imputation combined with regression calibration, full information maximum likelihood within a structural equation modeling framework, and a Bayesian model. In a simulation study, we assess the performance of the four approaches compared with more commonly used approaches that do not account for measurement error, missing values, or confounding.

RESULTS

The results demonstrate that the four approaches consistently outperform the alternative approaches on all performance metrics (bias, mean squared error, and confidence interval coverage). Even in simulated data of 100 subjects, these approaches perform well.

CONCLUSION

There can be a large benefit of addressing measurement error, missing values, and confounding to improve the estimation of exposure-outcome relations, even when the available sample size is relatively small.

Physical Frailty Phenotype Criteria and Their Synergistic Association on Cognitive Functioning

BACKGROUND

Frailty (physical frailty phenotype [PFP]) and its criteria (slow gait, weakness, weight loss, low activity, and exhaustion) are each associated with cognitive dysfunction. The extent to which the PFP is associated with cognition beyond that expected from its component parts remains uncertain.

METHOD

We used the National Health and Aging Trends Study to quantify associations between PFP criteria and cognitive performance (level/change) using adjusted mixed effects models. We tested whether frailty was associated with excess cognitive vulnerability (synergistic/excess effects, Cohen's d) beyond criteria contributions by assessing interactions between each criterion and frailty.

RESULTS

Among 7439 community-dwelling older adults (mean age = 75.2 years) followed for a mean of 3.2 years (SE = 0.03), 14.1% were frail. The PFP and PFP criteria were all associated with lower baseline cognitive performance, among which slow gait (-0.31 SD, SE = 0.02) and frailty (-0.23 SD, SE = 0.02) were strongest. Only slow gait (-0.03 SD/year, SE = 0.01), frailty (-0.02 SD/year, SE = 0.01), weight loss (-0.02 SD/year, SE = 0.01), and weakness (-0.02 SD/year, SE = 0.01) were associated with cognitive decline. Frailty was associated with cognitive performance above and beyond each criterion (excess effects ranging from -0.07 SD [SE = -0.05] for slow gait to -0.23 SD [SE = 0.03] for weakness); the same was not true for cognitive decline. Slow gait was the only criterion associated with cognitive change among both frail and nonfrail participants (frail: Cohen's d/year = -0.03, SE = 0.01; nonfrail: Cohen's d/year = -0.02, SE = 0.01).

CONCLUSIONS

PFP is an important frailty measure that is cross-sectionally associated with lower cognitive performance, but not with subsequent cognitive decline, above and beyond its criteria contributions. Further research into the construct of frailty as a "syndrome" correlated with cognition and other adverse outcomes is needed.

Using Bayesian networks to clarify interpretation of exposure-response regression coefficients: blood lead-mortality association as an example

We examine how Bayesian network (BN) learning and analysis methods can help to meet several methodological challenges that arise in interpreting significant regression coefficients in exposure-response regression modeling. As a motivating example, we consider the challenge of interpreting positive regression coefficients for blood lead level (BLL) as a predictor of mortality risk for nonsmoking men. We first note that practices such as dichotomizing or categorizing continuous confounders (e.g. income), omitting potentially important socioeconomic confounders (e.g. education), and assuming specific parametric regression model forms leave unclear to what extent a positive regression coefficient reflects these modeling choices, rather than a direct dependence of mortality risk on exposure. Therefore, significant exposure-response coefficients in parametric regression models do not necessarily reveal the extent to which reducing exposure-related variables (e.g. BLL) alone, while leaving fixed other correlates of exposure and mortality risks (e.g. education, income, etc.) would reduce adverse outcome risks (e.g. mortality risks). We then consider how BN structure-learning and inference algorithms and nonparametric estimation methods (partial dependence plots) can be used to clarify dependencies between variables, variable selection, confounding, and quantification of joint effects of multiple factors on risk, including possible high-order interactions and nonlinearities. We conclude that these details must be carefully modeled to determine whether a data set provides evidence that exposure itself directly affects risks; and that BN and nonparametric effect estimation and uncertainty quantification methods can complement regression modeling and help to improve the scientific basis for risk management decisions and policy-making by addressing these issues.

Effect of Sample Storage Conditions on Measurements of Salivary Cotinine Levels

Information on smoking exposure obtained with self-reports may be inaccurate. Cotinine has a large half-life and its salivary levels correlate well with plasmatic levels. The influence of storage conditions on the validity and precision of salivary cotinine assessments has rarely been evaluated. Here, smokers donated saliva samples, which were sent for immediate analysis, mail posting, storage at 4 °C for 30 or 90 days, or storage at −20 °C for 30 or 90 days. Cotinine levels were determined using enzyme-linked immune-sorbent assay. Agreement of cotinine level measurements was assessed using Bland-Altman analyses. Average age (years), duration of smoking (years) and number of cigarettes smoked (/day) were 55.4 (±SD 9.4), 35.1 (±SD 11.3), and 15.3 (±SD 7.6). The mean immediate cotinine level was 457 ng/mL (range 11.3 to 1318 ng/mL). Mean cotinine levels in samples analyzed after delay ranged between 433 ng/mL (−20 °C 30 days) and 468 ng/mL (4 °C 30 days). A dose-response gradient was observed in the relationship between salivary cotinine level and self-reported smoking status. A good agreement between cotinine levels for all storage conditions compared with immediate analysis was observed, with average differences ranging from −11 to 24 ng/mL. Cotinine levels remained stable regardless of the tested condition. The stability of salivary cotinine may enable samples to be obtained in difficult-to-reach areas, reduce study costs, and improve the validity of the information on exposure to smoking.

Implications of nonlinearity, confounding, and interactions for estimating exposure concentration-response functions in quantitative risk analysis

Recent advances in understanding of biological mechanisms and adverse outcome pathways for many exposure-related diseases show that certain common mechanisms involve thresholds and nonlinearities in biological exposure concentration-response (C-R) functions. These range from ultrasensitive molecular switches in signaling pathways, to assembly and activation of inflammasomes, to rupture of lysosomes and pyroptosis of cells. Realistic dose-response modeling and risk analysis must confront the reality of nonlinear C-R functions. This paper reviews several challenges for traditional statistical regression modeling of C-R functions with thresholds and nonlinearities, together with methods for overcoming them. Statistically significantly positive exposure-response regression coefficients can arise from many non-causal sources such as model specification errors, incompletely controlled confounding, exposure estimation errors, attribution of interactions to factors, associations among explanatory variables, or coincident historical trends. If so, the unadjusted regression coefficients do not necessarily predict how or whether reducing exposure would reduce risk. We discuss statistical options for controlling for such threats, and advocate causal Bayesian networks and dynamic simulation models as potentially valuable complements to nonparametric regression modeling for assessing causally interpretable nonlinear C-R functions and understanding how time patterns of exposures affect risk. We conclude that these approaches are promising for extending the great advances made in statistical C-R modeling methods in recent decades to clarify how to design regulations that are more causally effective in protecting human health.

Does Older Age have an Impact on Rituximab Efficacy and Safety? Results from the NOR-DMARD Register

OBJECTIVE

The objective of this study was to compare the efficacy and safety of rituximab in older vs younger patients with rheumatoid arthritis.

METHODS

Data on 367 patients with rheumatoid arthritis treated with rituximab in the Norwegian Disease-Modifying Antirheumatic Drug (NOR-DMARD) register were analysed, comparing patients aged ≥ 65 years (n = 91) with patients aged < 65 years (n = 276). Drug survival was compared using a Kaplan-Meier analysis and Cox proportional hazard models. Disease activity, as assessed by the Disease Activity Score based on 28 joints and erythrocyte sedimentation rate (DAS28-ESR) and the Simplified Disease Activity Index, was analysed with linear mixed models. The occurrence of adverse events was analysed by quasi-Poisson regression models.

RESULTS

Drug survival was similar in the two age groups. The proportion of patients who remained taking rituximab over 2 years was 72% in those under aged 65 years vs 74% in those aged ≥ 65 years. No statistically significant association with age was found for drug survival in either the unadjusted (hazard ratio 1.13, p = 0.65) or adjusted Cox proportional hazard analyses for the model with DAS28-ESR as a confounder (effect size 1.11, p = 0.73). Models including the Simplified Disease Activity Index instead of DAS28-ESR yielded similar results. Age was furthermore not significantly associated with disease activity over time, although there was a tendency towards a poorer response in older patients. In the older age group, there was a higher incidence of pneumonia (107 vs 51 per 1000 patient-years) and other serious infections (142 vs 66 per 1000 patient-years).

CONCLUSIONS

Rituximab is a reasonable therapeutic option for older patients with rheumatoid arthritis although vigilance is needed with regard to the infection profile.

TRIAL REGISTRATION

ClinicalTrials.gov Identifier: NCT01581294.

Confounding adjustment performance of ordinal analysis methods in stroke studies

Background

In stroke studies, ordinal logistic regression (OLR) is often used to analyze outcome on the modified Rankin Scale (mRS), whereas the non-parametric Mann-Whitney measure of superiority (MWS) has also been suggested. It is unclear how these perform comparatively when confounding adjustment is warranted.

Aims

Our aim is to quantify the performance of OLR and MWS in different confounding variable settings.

Methods

We set up a simulation study with three different scenarios; (1) dichotomous confounding variables, (2) continuous confounding variables, and (3) confounding variable settings mimicking a study on functional outcome after stroke. We compared adjusted ordinal logistic regression (aOLR) and stratified Mann-Whitney measure of superiority (sMWS), and also used propensity scores to stratify the MWS (psMWS). For comparability, OLR estimates were transformed to a MWS. We report bias, the percentage of runs that produced a point estimate deviating by more than 0.05 points (point estimate variation), and the coverage probability.

Results

In scenario 1, there was no bias in both sMWS and aOLR, with similar point estimate variation and coverage probabilities. In scenario 2, sMWS resulted in more bias (0.04 versus 0.00), and higher point estimate variation (41.6% versus 3.3%), whereas coverage probabilities were similar. In scenario 3, there was no bias in both methods, point estimate variation was higher in the sMWS (6.7%) versus aOLR (1.1%), and coverage probabilities were 0.98 (sMWS) versus 0.95 (aOLR). With psMWS, bias remained 0.00, with less point estimate variation (1.5%) and a coverage probability of 0.95.

Conclusions

The bias of both adjustment methods was similar in our stroke simulation scenario, and the higher point estimate variation in the MWS improved with propensity score based stratification. The stratified MWS is a valid alternative for adjusted OLR only when the ratio of number of strata versus number of observations is relatively low, but propensity score based stratification extends the application range of the MWS.

State of the art in selection of variables and functional forms in multivariable analysis-outstanding issues

BACKGROUND

How to select variables and identify functional forms for continuous variables is a key concern when creating a multivariable model. Ad hoc 'traditional' approaches to variable selection have been in use for at least 50 years. Similarly, methods for determining functional forms for continuous variables were first suggested many years ago. More recently, many alternative approaches to address these two challenges have been proposed, but knowledge of their properties and meaningful comparisons between them are scarce. To define a state of the art and to provide evidence-supported guidance to researchers who have only a basic level of statistical knowledge, many outstanding issues in multivariable modelling remain. Our main aims are to identify and illustrate such gaps in the literature and present them at a moderate technical level to the wide community of practitioners, researchers and students of statistics.

METHODS

We briefly discuss general issues in building descriptive regression models, strategies for variable selection, different ways of choosing functional forms for continuous variables and methods for combining the selection of variables and functions. We discuss two examples, taken from the medical literature, to illustrate problems in the practice of modelling.

RESULTS

Our overview revealed that there is not yet enough evidence on which to base recommendations for the selection of variables and functional forms in multivariable analysis. Such evidence may come from comparisons between alternative methods. In particular, we highlight seven important topics that require further investigation and make suggestions for the direction of further research.

CONCLUSIONS

Selection of variables and of functional forms are important topics in multivariable analysis. To define a state of the art and to provide evidence-supported guidance to researchers who have only a basic level of statistical knowledge, further comparative research is required.

Evaluation of confounding in epidemiologic studies assessing alcohol consumption on the risk of ischemic heart disease

Background

Among different investigators studying the same exposures and outcomes, there may be a lack of consensus about potential confounders that should be considered as matching, adjustment, or stratification variables in observational studies. Concerns have been raised that confounding factors may affect the results obtained for the alcohol-ischemic heart disease relationship, as well as their consistency and reproducibility across different studies. Therefore, we assessed how confounders are defined, operationalized, and discussed across individual studies evaluating the impact of alcohol on ischemic heart disease risk.

Methods

For observational studies included in a recent alcohol-ischemic heart disease meta-analysis, we identified all variables adjusted, matched, or stratified for in the largest reported multivariate model (i.e. potential confounders). We recorded how the variables were measured and grouped them into higher-level confounder domains. Abstracts and Discussion sections were then assessed to determine whether authors considered confounding when interpreting their study findings.

Results

85 of 87 (97.7%) studies reported multivariate analyses for an alcohol-ischemic heart disease relationship. The most common higher-level confounder domains included were smoking (79, 92.9%), age (74, 87.1%), and BMI, height, and/or weight (57, 67.1%). However, no two models adjusted, matched, or stratified for the same higher-level confounder domains. Most (74/87, 85.1%) articles mentioned or alluded to “confounding” in their Abstract or Discussion sections, but only one stated that their main findings were likely to be affected by residual confounding. There were five (5/87, 5.7%) authors that explicitly asked for caution when interpreting results.

Conclusion

There is large variation in the confounders considered across observational studies evaluating the impact of alcohol on ischemic heart disease risk and almost all studies spuriously ignore or eventually dismiss confounding in their conclusions. Given that study results and interpretations may be affected by the mix of potential confounders included within multivariate models, efforts are necessary to standardize approaches for selecting and accounting for confounders in observational studies.

Radiation Therapy Practice Patterns for Brain Metastases in the United States in the Stereotactic Radiosurgery Era

PURPOSE

Utilization of stereotactic radiosurgery (SRS) for brain metastases (BM) has increased, prompting reassessment of whole brain radiation therapy (WBRT). A pattern of care analysis of SRS and WBRT dose-fractionations was performed in patients presenting with BM at the time of cancer diagnosis.

METHODS AND MATERIALS

Adults with BM at cancer diagnosis between 2010 to 2015 and no prior malignancy were identified in the National Cancer Database. SRS was defined using published thresholds. Short (ShWBRT), standard (StWBRT), and extended (ExWBRT) dose-fractionations were defined as 4 to 9, 10 to 15, and >15 fractions. Radioresistant histology was defined as melanoma, renal cell carcinoma, sarcoma or spindle cell, or gastrointestinal primary.

RESULTS

Of 4,087,967 adults with their first lifetime cancer, 90,388 (2.2%) had BM at initial diagnosis. Of these, 11,486 (12.7%) received SRS and 24,262 (26.8%) WBRT as first-course radiation therapy. The proportion of annual WBRT use decreased from 27.8% to 23.5% of newly diagnosed patients, and SRS increased from 8.7% to 17.9%. Common dose-fractionations were 30 Gy in 10 fractions (56.8%) for WBRT and 20 Gy in 1 fraction (13.0%) for SRS. On multivariate analysis, factors significantly associated with SRS versus WBRT included later year of diagnosis (2015 vs 2010, adjusted odds ratio [aOR] = 2.4), radioresistance (aOR = 2.0), academic facility (aOR = 1.9), highest income quartile (aOR = 1.6), chemotherapy administration (aOR = 1.4), and longer travel distance (>15 vs < 5 miles, aOR = 1.4). Linear regression revealed significant ExWBRT reductions (-22.4%/y, R2 = 0.97, P < .001) and no significant change for ShWBRT or StWBRT. Patients were significantly more likely to receive ShWBRT than StWBRT if not treated with chemotherapy (aOR = 3.5).

CONCLUSIONS

Utilization of WBRT, particularly ExWBRT, decreased while SRS utilization doubled as the first radiation therapy course in patients with BM at diagnosis. Patients with radioresistant histologies were more likely to receive SRS. Those not receiving chemotherapy, potentially owing to poor performance status, were less likely to receive SRS and more likely to receive ShWBRT.

Association of Diabetes Mellitus and Biomarkers of Abnormal Glucose Metabolism With Incident Radiographic Knee Osteoarthritis

OBJECTIVE

The association of diabetes mellitus (DM) with increased risk of knee osteoarthritis (OA) is uncertain. We evaluated associations of DM and biomarkers of abnormal glucose metabolism with incident radiographic knee OA, controlling for body mass index (BMI).

METHODS

Participants (mean ± SD age 60.6 ± 7.8 years; mean ± SD body mass index [BMI] 29.1 ± 4.9 kg/m2 ) were from the Multicenter Osteoarthritis Study and did not have radiographic knee OA at baseline (Kellgren/Lawrence [K/L] grade <2 bilaterally). A random sample (n = 987) was selected and stratified by BMI. Baseline serum fasting glucose and homeostasis model assessment-estimated insulin resistance (HOMA-IR) were measured. Participants were categorized as having DM based on self-report, use of medication, or fasting glucose ≥126 mg/dl. Incident radiographic knee OA (K/L grade ≥2 or knee replacement) was assessed at 3 follow-up visits (30, 60, and 84 months). Knee-level pooled logistic regression analysis was performed to obtain odds ratios (ORs) (95% confidence interval [95% CI]) for associations of DM status and biomarkers of abnormal glucose metabolism with incident radiographic knee OA.

RESULTS

After adjustment for BMI, the odds of incident radiographic knee OA were not associated with baseline DM status nor with levels of fasting glucose and HOMA-IR, overall and in men. In women, HOMA-IR was inversely associated with odds of incident radiographic knee OA (adjusted OR 0.80 [95% CI 0.69-0.94], P = 0.005).

CONCLUSION

DM and higher levels of biomarkers of abnormal glucose metabolism were not associated with increased odds of incident radiographic knee OA after adjusting for BMI in this cohort overall. A possible protective association of higher HOMA-IR with incident radiographic knee OA in women warrants further investigation.

The oral glucose tolerance test-derived incremental glucose peak is associated with greater arterial stiffness and maladaptive arterial remodeling: The Maastricht Study

Background

Daily glucose variability may contribute to vascular complication development irrespective of mean glucose values. The incremental glucose peak (IGP) during an oral glucose tolerance test (OGTT) can be used as a proxy of glucose variability. We investigated the association of IGP with arterial stiffness, arterial remodeling, and microvascular function, independent of HbA_1c and other confounders.

Methods

IGP was calculated as the peak minus baseline plasma glucose value during a seven-point OGTT in 2758 participants (age: 60 ± 8 years; 48% women) of The Maastricht Study, an observational population-based cohort. We assessed the cross-sectional associations between IGP and arterial stiffness (carotid-femoral pulse wave velocity [cf-PWV], carotid distensibility coefficient [carDC]), arterial remodeling (carotid intima-media thickness [cIMT]; mean [CWS_mean] and pulsatile [CWS_puls] circumferential wall stress), and microvascular function (retinal arteriolar average dilatation; heat-induced skin hyperemia) via multiple linear regression with adjustment for age, sex, HbA_1c, cardiovascular risk factors, lifestyle factors, and medication use.

Results

Higher IGP was independently associated with higher cf-PWV (regression coefficient [B]: 0.054 m/s [0.020; 0.089]) and with higher CWS_mean (B: 0.227 kPa [0.008; 0.446]). IGP was not independently associated with carDC (B: − 0.026 10⁻³/kPa [− 0.112; 0.060]), cIMT (B: − 2.745 µm [− 5.736; 0.245]), CWS_puls (B: 0.108 kPa [− 0.054; 0.270]), retinal arteriolar average dilatation (B: − 0.022% [− 0.087; 0.043]), or heat-induced skin hyperemia (B: − 1.380% [− 22.273; 19.513]).

Conclusions

IGP was independently associated with aortic stiffness and maladaptive carotid remodeling, but not with carotid stiffness, cIMT, and microvascular function measures. Future studies should investigate whether glucose variability is associated with cardiovascular disease.

Evaluating Flexible Modeling of Continuous Covariates in Inverse-Weighted Estimators

Correct specification of the exposure model is essential for unbiased estimation in marginal structural models with inverse-probability-of-treatment weights. However, although flexible modeling is commonplace when estimating effects of continuous covariates in outcome models, its use is less frequent in estimation of inverse probability weights. Using simulations, we assess the accuracy of the treatment effect estimates and covariate balance obtained with different exposure model specifications when the true relationship between a continuous, possibly time-varying covariate Lt and the logit of the probability of exposure is nonlinear. Specifically, we compare 4 approaches to modeling the effect of Lt when estimating inverse probability weights: a linear function, the covariate-balancing propensity score, and 2 easy-to-implement flexible methods that relax the assumption of linearity: cubic regression splines and fractional polynomials. Using data from 2 empirical studies, we compare linear exposure models with flexible exposure models to estimate the effect of sustained virological response to hepatitis C virus treatment on the progression of liver fibrosis. Our simulation results demonstrate that ignoring important nonlinear relationships when fitting the exposure model may provide poorer covariate balance and induce substantial bias in the estimated exposure-outcome associations. Analysts should routinely consider flexible modeling of continuous covariates when estimating inverse-probability-of-treatment weights.

National Rate of Tobacco and Substance Use Disorders Among Hospitalized Heart Failure Patients

BACKGROUND

Several cardiotoxic substances impact heart failure incidence. The burden of comorbid tobacco or substance use disorders among heart failure patients is under-characterized. We describe the burden of tobacco and substance use disorders among hospitalized heart failure patients in the United States.

METHODS

We calculated the proportion of primary heart failure hospitalizations in the 2014 National Inpatient Sample with tobacco or substance use disorders accounting for demographic factors.

RESULTS

Of 989,080 heart failure hospitalizations, 15.5% (n = 152,965) had documented tobacco (n = 119,285, 12.1%) or substance (n = 61,510, 6.2%) use disorder. Female sex was associated with lower rates of tobacco (odds ratio [OR] 0.72; 95% confidence interval [CI], 0.70-0.74) and substance (OR 0.37; 95% CI, 0.36-0.39) use disorder. Tobacco and substance use disorder rates were highest for hospitalizations <55years of age. Native American race was associated with increased risk of alcohol use disorder (OR 1.67; 95% CI, 1.27-2.20) and black race with alcohol (OR 1.09; 95% CI, 1.02-1.16) or drug (OR 1.63; 95% CI, 1.53-1.74) use disorder. Medicaid insurance or income in the lowest quartile were associated with increased risk of tobacco and substance use disorders.

CONCLUSIONS

Tobacco and substance use disorders affect vulnerable heart failure populations, including those of male sex, younger age, lower socioeconomic status, and racial/ethnic minorities. Enhanced screening for tobacco and substance use disorders in hospitalized heart failure patients may reveal opportunities for treatment and secondary prevention.

Factors Related to the Assessment of Resting Metabolic Rate in Critically Ill Patients

BACKGROUND

Predicting resting metabolic rate (RMR) in mechanically ventilated, critically ill patients is an important part of the nutrition care in such patients.

METHODS

RMR and associated clinical data from various studies of mechanically ventilated, critically ill patients were combined, and the impact of body size, age, reason for admission, and sedation level were analyzed along with prediction methods of RMR (the American Society for Parenteral and Enteral Nutrition [ASPEN] standards and the Penn State equation).

RESULTS

Among 826 measurements, trauma patients had a higher RMR than surgical and medical patients (2077 ± 290 vs 1987 ± 282 kcal/d; P < .0001). RMR was not different in sedated vs unsedated patients. Wide ranges of weight (27-374 kg) and age (18-95 years) were captured. The relationships between weight and RMR and RMR and age were curvilinear. For weight-based ratio methods of RMR prediction, <50% of predictions were within the range in which they were designed to work. The accuracy of the Penn State equation was better in some weight categories than others. New equations based on a wider range of body weights and ages are presented.

CONCLUSIONS

Curvilinear functions exist for weight and age in relation to RMR, but extraordinary levels of each are required for the curve to become apparent. The ASPEN energy standards (kcal/kg body weight) fail to predict RMR because the relationship is more complex than a simple ratio. The Penn State equations are better able to model these relationships. The new versions of the equation presented here await validation.

Duvalier Regime in Haiti and Immigrant Health in the United States

BACKGROUND

Haitians immigrate to the United States for many reasons, including the opportunity to escape political violence. The extant literature on Haitian immigrant health focuses on post-migration, rather than pre-migration, environments and experiences. Objective: In this study, we analyze health outcomes data from a nationally representative sample of Haitian immigrants in the United States from 1996 to 2015. We estimate age-adjusted associations between pre-migration residence in Haiti during the repressive regimes and generalized terror of Francois and Jean-Claude Duvalier, who ran Haiti from 1957 to 1986.

METHODS

We used ordered probit regression models to quantify age-adjusted associations between the duration of pre-migration residence in Haiti during the Duvalier regime, and the distribution of post-migration health status among Haitian immigrants in the United States. Findings: Our study sample included 2,438 males and 2,800 females ages 15 and above. The mean age of males was 43.5 (standard deviation, 15.5) and the mean age of females was 44.7 (standard deviation, 16.6). Each additional decade of pre-migration residence in Haiti during the Duvalier regime is associated with a 2.9 percentage point decrease (95% confidence interval 0.6 to 5.3) in excellent post-migration health for males, and a 2.8 percentage point decrease (95% confidence interval, 0.8 to 4.8) for females. Within the subsample of Haitian immigrants with any pre-migration residence in Haiti during the Duvalier regime, each additional decade since the regime is associated with a 3.3 percentage point increase (95% confidence interval, 1.2 to 5.5) in excellent post-migration health for males, and a 2.3 percentage point increase (95% confidence interval, 0.5 to 4.1) for females.

CONCLUSIONS

Overall, we found statistically significant and negative associations between the Duvalier regime and the post-migration distribution of health status 10 to 57 years later. We found statistically significant and positive associations between the length of time since the Duvalier regime and post-migration health.

Duvalier Regime in Haiti and Immigrant Health in the United States

Background:

Haitians immigrate to the United States for many reasons, including the opportunity to escape political violence. The extant literature on Haitian immigrant health focuses on post-migration, rather than pre-migration, environments and experiences.

Objective:

In this study, we analyze health outcomes data from a nationally representative sample of Haitian immigrants in the United States from 1996 to 2015. We estimate age-adjusted associations between pre-migration residence in Haiti during the repressive regimes and generalized terror of Francois and Jean-Claude Duvalier, who ran Haiti from 1957 to 1986.

Methods:

We used ordered probit regression models to quantify age-adjusted associations between the duration of pre-migration residence in Haiti during the Duvalier regime, and the distribution of post-migration health status among Haitian immigrants in the United States.

Findings:

Our study sample included 2,438 males and 2,800 females ages 15 and above. The mean age of males was 43.5 (standard deviation, 15.5) and the mean age of females was 44.7 (standard deviation, 16.6). Each additional decade of pre-migration residence in Haiti during the Duvalier regime is associated with a 2.9 percentage point decrease (95% confidence interval 0.6 to 5.3) in excellent post-migration health for males, and a 2.8 percentage point decrease (95% confidence interval, 0.8 to 4.8) for females. Within the subsample of Haitian immigrants with any pre-migration residence in Haiti during the Duvalier regime, each additional decade since the regime is associated with a 3.3 percentage point increase (95% confidence interval, 1.2 to 5.5) in excellent post-migration health for males, and a 2.3 percentage point increase (95% confidence interval, 0.5 to 4.1) for females.

Conclusions:

Overall, we found statistically significant and negative associations between the Duvalier regime and the post-migration distribution of health status 10 to 57 years later. We found statistically significant and positive associations between the length of time since the Duvalier regime and post-migration health.

Low fitness is associated with abdominal adiposity and low-grade inflammation independent of BMI

OBJECTIVE

Up to 30% of obese individuals are metabolically healthy. Metabolically healthy obese (MHO) individuals are characterized by having low abdominal adiposity, low inflammation level and low risk of developing metabolic comorbidity. In this study, we hypothesize that cardiorespiratory fitness (fitness) is a determinant factor for the MHO individuals and aim to investigate the associations between fitness, abdominal adiposity and low-grade inflammation within different BMI categories.

METHOD

Data from 10,976 individuals from the general population, DANHES 2007-2008, on waist circumference, fitness and C-reactive protein (hsCRP) were analysed using multiple linear and median quantile regressions.

RESULTS

In men, an inverse association between fitness (+5 mL min-1 kg-1) and waist circumference (-1.45 cm; 95% CI: -1.55 to -1.35 cm; p<0.001), and an inverse association between fitness (+5 mL min-1 kg-1) and hsCRP (-0.22 mg/L; 95% CI: -0.255 to -0.185 mg/L; p<0.001) was found, all independent of BMI. Similarly in women, an inverse association between fitness (+5 mL min-1 kg-1) and waist circumference (-1.15 cm; 95% CI: -1.25 to -1.0 cm; p<0.001), and an inverse association between fitness (+5 mL min-1 kg-1) and hsCRP (-0.26 mg/L; 95% CI: -0.3 to -0.22 mg/L; p<0.001) was found, all independent of BMI. Additionally, significant positive associations between waist circumference and hsCRP were found for both men and women, independently of BMI.

CONCLUSION

Fitness was found to be inversely associated with both abdominal adiposity and low-grade inflammation independent of BMI. These data suggest that, in spite of BMI, high fitness levels lead to a reduction in abdominal fat mass and low-grade inflammation.

The Influence of Sex in Stroke Thrombolysis: A Systematic Review and Meta-Analysis

Background and Purpose

There is increasing recognition of the importance of stroke in females to both clinical and public health. The natural course of stroke is worse in females than in males, but the evidence regarding sex disparities in the responses to thrombolysis in stroke patents is still controversial. We compared outcomes after thrombolysis treatment between females and males.

Methods

Clinical trials reported in the Embase, PubMed, and Cochrane Library electronic databases up to March 13, 2017 were included in this analysis. Two reviewers independently extracted the data and conducted quality assessments. Statistical tests were performed to check for heterogeneity and publication bias. Sensitivity analysis was also performed to evaluate the stability of the conclusions.

Results

Sixteen reports involving 60,159 patients were available for analysis. The female patients were a 0.89-fold [95% confidence interval (CI)=0.87–0.90, p<0.001], 0.89-fold (95% CI=0.87–0.91, p<0.001), and 1.24-fold (95% CI=1.11–1.36, p<0.001) more likely to obtain good, excellent, and poor functional outcomes, respectively, with no significant difference in the complications of symptomatic intracranial hemorrhage among the sexes [risk ratios (RR)=0.99, 95% CI=0.92–1.07, p=0.81] after thrombolysis treatment. In addition, the prevalence of a good functional outcome did not differ significantly between females and males in the intra-arterial thrombolysis (IAT) group (RR=1.05, 95% CI=0.85–1.29, p=0.67) in a subgroup analysis.

Conclusions

This study has demonstrated that females often exhibit a worse outcome than males after intravenous thrombolysis (IVT), whereas no relevant sex differences were found in outcome or recanalization after IAT, with safety regarding hemorrhage complications from thrombolysis being the same for the sexes. However, IVT should not be withheld from female stroke patients solely based on their sex before the findings are confirmed in further large-scale research.

Estimating the effects of second-line therapy for type 2 diabetes mellitus: retrospective cohort study

OBJECTIVE

Metformin is the recommended initial drug treatment in type 2 diabetes mellitus, but there is no clearly preferred choice for an additional drug when indicated. We compare the counterfactual drug effectiveness in lowering glycated hemoglobin (HbA1c) levels and effect on body mass index (BMI) of four diabetes second-line drug classes using electronic health records.

STUDY DESIGN AND SETTING

Retrospective analysis of electronic health records of US-based patients in the Explorys database using causal inference methodology to adjust for patient censoring and confounders.

PARTICIPANTS AND EXPOSURES

Our cohort consisted of more than 40 000 patients with type 2 diabetes, prescribed metformin along with a drug out of four second-line drug classes-sulfonylureas, thiazolidinediones, dipeptidyl peptidase 4 (DPP-4) inhibitors and glucagon-like peptide-1 agonists-during the years 2000-2015. Roughly, 17 000 of these patients were followed for 12 months after being prescribed a second-line drug.

MAIN OUTCOME MEASURES

HbA1c and BMI of these patients after 6 and 12 months following treatment.

RESULTS

We demonstrate that all four drug classes reduce HbA1c levels, but the effect of sulfonylureas after 6 and 12 months of treatment is less pronounced compared with other classes. We also estimate that DPP-4 inhibitors decrease body weight significantly more than sulfonylureas and thiazolidinediones.

CONCLUSION

Our results are in line with current knowledge on second-line drug effectiveness and effect on BMI. They demonstrate that causal inference from electronic health records is an effective way for conducting multitreatment causal inference studies.