Pooled Analysis Of Individual-level Data Research Articles

Covariate balance-related propensity score weighting in estimating overall hazard ratio with distributed survival data

BackgroundWhen data is distributed across multiple sites, sharing information at the individual level among sites may be difficult. In these multi-site studies, propensity score model can be fitted with data within each site or data from all sites when using inverse probability-weighted Cox regression to estimate overall hazard ratio. However, when there is unknown heterogeneity of covariates in different sites, either approach may lead to potential bias or reduced efficiency. In this study, we proposed a method to estimate propensity score based on covariate balance-related criterion and estimate the overall hazard ratio while overcoming data sharing constraints across sites.MethodsThe proposed propensity score was generated by choosing between global and local propensity score based on covariate balance-related criterion, combining the global propensity score fitted in the entire population and the local propensity score fitted within each site. We used this proposed propensity score to estimate overall hazard ratio of distributed survival data with multiple sites, while requiring only the summary-level information across sites. We conducted simulation studies to evaluate the performance of the proposed method. Besides, we applied the proposed method to real-world data to examine the effect of radiation therapy on time to death among breast cancer patients.ResultsThe simulation studies showed that the proposed method improved the performance in estimating overall hazard ratio comparing with global and local propensity score method, regardless of the number of sites and sample size in each site. Similar results were observed under both homogeneous and heterogeneous settings. Besides, the proposed method yielded identical results to the pooled individual-level data analysis. The real-world data analysis indicated that the proposed method was more likely to find a significant effect of radiation therapy on mortality compared to the global propensity score method and local propensity score method.ConclusionsThe proposed covariate balance-related propensity score in multi-site distributed survival data outperformed the global propensity score estimated using data from the entire population or the local propensity score estimated within each site in estimating the overall hazard ratio. The proposed approach can be performed without individual-level data transfer between sites and would yield the same results as the corresponding pooled individual-level data analysis.

Impact of body mass index and its change on survival outcomes in patients with early breast cancer: A pooled analysis of individual-level data from BCIRG-001 and BCIRG-005 trials.

The relationships between body mass index (BMI) and survival rates are complex, and have not been thoroughly investigated in breast cancer patients who received adjuvant chemotherapy. We collected data on 2394 patients from two randomized, phase III clinical trials that investigated adjuvant chemotherapy in breast cancer identified in Project Data Sphere. The objective was to examine the effect of baseline BMI, BMI after adjuvant chemotherapy, and BMI change from baseline to post-adjuvant chemotherapy on disease-free survival (DFS) and overall survival (OS). Restricted cubic splines were used to examine potential non-linear associations between continuous BMI value and survival. Stratified analyses involved chemotherapy regimens. Severe obesity (BMI≥40.0kg/m2) at baseline was independently associated with worse DFS (hazard ration [HR]=1.48, 95% confidence interval [CI] 1.02-2.16, P=0.04) and OS (HR=1.79, 95%CI 1.17-2.74, P=0.007) compared with underweight/normal weight (BMI≤24.9kg/m2). A BMI loss >10% was also an independent prognostic factor for adverse OS (HR=2.14, 95%CI 1.17-3.93, P=0.014). Stratified analyses revealed that severe obesity adversely affected DFS (HR=2.38, 95%CI 1.26-4.34, P=0.007) and OS (HR=2.90, 95%CI 1.46-5.76, P=0.002) in the docetaxel-based group, but not in the non-docetaxel-based group. Restricted cubic splines revealed a "J-shaped" association of baseline BMI with risk of recurrence or all-cause death, and this relationship was more pronounced in the docetaxel-based group. In early breast cancer patients treated with adjuvant chemotherapy, baseline severe obesity was significantly linked to worse DFS and OS, and a BMI loss over 10% from baseline to post-adjuvant chemotherapy also negatively affected OS. Moreover, the prognostic role of BMI might differ between docetaxel-based and non-docetaxel-based groups.

239MO Impact of body mass index and its change on survival outcomes in patients with early breast cancer: A pooled analysis of individual-level data from BCIRG-001 and BCIRG-005 trials

239MO Impact of body mass index and its change on survival outcomes in patients with early breast cancer: A pooled analysis of individual-level data from BCIRG-001 and BCIRG-005 trials

Patiromer Treatment in Patients With CKD, Hyperkalemia, and Hyperphosphatemia: A Post Hoc Analysis of 3 Clinical Trials

Patients with chronic kidney disease (CKD), hyperkalemia (serum potassium [sK+]>5.0 mEq/L), and hyperphosphatemia experience poor clinical outcomes. Patiromer, a potassium binder that uses calcium as the exchange ion, may also reduce serum phosphorus (sP). We characterized the effect ofpatiromer on sP in patients with CKD, hyperkalemia, and hyperphosphatemia. A post hoc pooled analysis of individual-level data from the AMETHYST-DN, OPAL-HK, and TOURMALINE trials of patiromer. Patients with CKD and hyperkalemia. Patients treated with patiromer (8.4-33.6 g/day). Mean changes from baseline in sP, sK+, serum calcium (sCa2+), and serum magnesium (sMg2+) after 2 and 4 weeks of treatment. Descriptive statistics to summarize pooled data on the study outcomes from the 3 studies. We included 578 patients in the analysis. Of these participants, 86 patients (14.9%) had baseline hyperphosphatemia of whom 75.6% (65 of 86) had CKD stage 4/5 and 31.1% (153 of 492) with sP≤4.5mg/dL had CKD stage4/5. Among the patients with elevated sPand sK+at baseline, the mean±SD reduction insP and sK+after 4 weeks of patiromer treatment was-0.62±1.09mg/dL and-0.71± 0.51mEq/L,respectively. Additionally, the mean±SDreduction in sMg2+in these patients was -0.25±0.23mg/dL while sCa2+remained unchanged. Both sMg2+and sCa2+remained within the normal range. Patiromer was generally well tolerated, and no serious adverse events were considered related to patiromer. These were post hoc analyses, no placebo comparison was performed due to the design of the original studies, and the follow-up period was limited to 4 weeks. Reductions in sP and sK+to the normal range were observed after 2 weeks of patiromer treatment, and the reduction was sustained during 4 weeks of treatment among patients with non-dialysis-dependent CKD, hyperkalemia, and hyperphosphatemia. Future controlled trials are needed to establish if patiromer is useful to reduce both sK+and sP in hyperkalemic patients with CKD and hyperphosphatemia.

Micronutrient deficiencies among preschool-aged children and women of reproductive age worldwide: a pooled analysis of individual-level data from population-representative surveys

Micronutrient deficiencies compromise immune systems, hinder child growth and development, and affect human potential worldwide. Yet, to our knowledge, the only existing estimate of the global prevalence of micronutrient deficiencies is from over 30 years ago and is based only on the prevalence of anaemia. We aimed to estimate the global and regional prevalence of deficiency in at least one of three micronutrients among preschool-aged children (aged 6-59 months) and non-pregnant women of reproductive age (aged 15-49 years). In this pooled analysis, we reanalysed individual-level biomarker data for micronutrient status from nationally representative, population-based surveys. We used Bayesian hierarchical logistic regression to estimate the prevalence of deficiency in at least one of three micronutrients for preschool-aged children (iron, zinc, and vitamin A) and for non-pregnant women of reproductive age (iron, zinc, and folate), globally and in seven regions using 24 nationally representative surveys done between 2003 and 2019. We estimated the global prevalence of deficiency in at least one of three micronutrients to be 56% (95% uncertainty interval [UI] 48-64) among preschool-aged children, and 69% (59-78) among non-pregnant women of reproductive age, equivalent to 372 million (95% UI 319-425) preschool-aged children and 1·2 billion (1·0-1·4) non-pregnant women of reproductive age. Regionally, three-quarters of preschool-aged children with micronutrient deficiencies live in south Asia (99 million, 95% UI 80-118), sub-Saharan Africa (98 million, 83-113), or east Asia and the Pacific (85 million, 61-110). Over half (57%) of non-pregnant women of reproductive age with micronutrient deficiencies live in east Asia and the Pacific (384 million, 279-470) or south Asia (307 million, 255-351). We estimate that over half of preschool-aged children and two-thirds of non-pregnant women of reproductive age worldwide have micronutrient deficiencies. However, estimates are uncertain due to the scarcity of population-based micronutrient deficiency data. US Agency for International Development.

The socioeconomic gradient of alcohol use: an analysis of nationally representative survey data from 55 low-income and middle-income countries

SummaryBackgroundAlcohol is a leading risk factor for over 200 conditions and an important contributor to socioeconomic health inequalities. However, little is known about the associations between individuals’ socioeconomic circumstances and alcohol consumption, especially heavy episodic drinking (HED; ≥5 drinks on one occasion) in low-income or middle-income countries. We investigated the association between individual and household level socioeconomic status, and alcohol drinking habits in these settings.MethodsIn this pooled analysis of individual-level data, we used available nationally representative surveys—mainly WHO Stepwise Approach to Surveillance surveys—conducted in 55 low-income and middle-income countries between 2005 and 2017 reporting on alcohol use. Surveys from participants aged 15 years or older were included. Logistic regression models controlling for age, country, and survey year stratified by sex and country income groups were used to investigate associations between two indicators of socioeconomic status (individual educational attainment and household wealth) and alcohol use (current drinking and HED amongst current drinkers).FindingsSurveys from 336 287 participants were included in the analysis. Among males, the highest prevalence of both current drinking and HED was found in lower-middle-income countries (L-MICs; current drinking 49·9% [95% CI 48·7–51·2] and HED 63·3% [61·0–65·7]). Among females, the prevalence of current drinking was highest in upper-middle-income countries (U-MIC; 29·5% [26·1–33·2]), and the prevalence of HED was highest in low-income countries (LICs; 36·8% [33·6–40·2]). Clear gradients in the prevalence of current drinking were observed across all country income groups, with a higher prevalence among participants with high socioeconomic status. However, in U-MICs, current drinkers with low socioeconomic status were more likely to engage in HED than participants with high socioeconomic status; the opposite was observed in LICs, and no association between socioeconomic status and HED was found in L-MICs.InterpretationThe findings call for urgent alcohol control policies and interventions in LICs and L-MICs to reduce harmful HED. Moreover, alcohol control policies need to be targeted at socially disadvantaged groups in U-MICs.FundingDeutsche Forschungsgemeinschaft and the National Center for Advancing Translational Sciences of the US National Institutes of Health.

Cardiovascular Health in India - a Report Card from Three Urban and Rural Surveys of 22,144 Adults.

Background:Markers of ideal cardiovascular health (CVH) predict cardiovascular events. We estimated the prevalence of ideal CVH markers in two levels of cities and villages in India.Methods:We did pooled analysis of individual-level data from three cross sectional surveys of adults ≥ 30 years over 2010–14 (CARRS: Centre for cArdiometabolic Risk Reduction in South Asia; UDAY and Solan Surveillance Study) representing metropolitan cities; smaller cities and rural areas in diverse locations of India. We defined ideal CVH using modified American Heart Association recommendations: not smoking, ≥ 5 servings of fruits and vegetables (F&V), high physical activity (PA), body mass index (BMI) <25 Kg/m2, blood pressure (BP) <120/80 mm Hg, fasting plasma glucose (FPG) <100 mg/dl, and total cholesterol (TC) <200 mg/dL. We estimated (1) age-and sex-standardized prevalence of ideal CVH and (2) prevalence of good (≥6 markers), moderate (4–5), and poor CVH (≤3) adjusted for age, sex, education, and stratified by setting and asset tertiles.Results:Of the total 22,144 participants, the prevalence of ideal CVH markers were: not smoking (76.7% [95% CI 76.1, 77.2]), consumed ≥5 F&V (4.2% [3.9, 4.5]), high PA (67.5% [66.8, 68.2]), optimum BMI (59.6% [58.9, 60.3]), ideal BP (34.5% [33.9, 35.2]), FPG (65.8% [65.1, 66.5]) and TC (65.4% [64.7, 66.1]). The mean number of ideal CVH metrics was 3.7(95% CI: 3.7, 3.8). Adjusted prevalence of good, moderate, and poor CVH, varied across settings: metropolitan (3.9%, 41.0%, and 55.1%), smaller cities (7.8%, 49.2%, and 43%), and rural (10.4%, 60.9%, and 28.7%) and across asset tertiles: Low (11.0%, 55.9%, 33.1%), Middle (6.3%, 52.2%, 41.5%), and High (5.0%, 46.4%, 48.7%), respectively.Conclusion:Achievement of ideal CVH varied, with higher prevalence in rural and lower asset tertiles. Multi-sectoral and targeted policy and program actions are needed to improve CVH in diverse contexts in India.

Epidemiology of anal human papillomavirus infection and high-grade squamous intraepithelial lesions in 29 900 men according to HIV status, sexuality, and age: a collaborative pooled analysis of 64 studies.

SummaryBackgroundRobust age-specific estimates of anal human papillomavirus (HPV) and high-grade squamous intraepithelial lesions (HSIL) in men can inform anal cancer prevention efforts. We aimed to evaluate the age-specific prevalence of anal HPV, HSIL, and their combination, in men, stratified by HIV status and sexuality.MethodsWe did a systematic review for studies on anal HPV infection in men and a pooled analysis of individual-level data from eligible studies across four groups: HIV-positive men who have sex with men (MSM), HIV-negative MSM, HIV-positive men who have sex with women (MSW), and HIV-negative MSW. Studies were required to inform on type-specific HPV infection (at least HPV16), detected by use of a PCR-based test from anal swabs, HIV status, sexuality (MSM, including those who have sex with men only or also with women, or MSW), and age. Authors of eligible studies with a sample size of 200 participants or more were invited to share deidentified individual-level data on the above four variables. Authors of studies including 40 or more HIV-positive MSW or 40 or more men from Africa (irrespective of HIV status and sexuality) were also invited to share these data. Pooled estimates of anal high-risk HPV (HR-HPV, including HPV16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, and 68), and HSIL or worse (HSIL+), were compared by use of adjusted prevalence ratios (aPRs) from generalised linear models.FindingsThe systematic review identified 93 eligible studies, of which 64 contributed data on 29 900 men to the pooled analysis. Among HIV-negative MSW anal HPV16 prevalence was 1·8% (91 of 5190) and HR-HPV prevalence was 6·9% (345 of 5003); among HIV-positive MSW the prevalences were 8·7% (59 of 682) and 26·9% (179 of 666); among HIV-negative MSM they were 13·7% (1455 of 10 617) and 41·2% (3798 of 9215), and among HIV-positive MSM 28·5% (3819 of 13 411) and 74·3% (8765 of 11 803). In HIV-positive MSM, HPV16 prevalence was 5·6% (two of 36) among those age 15–18 years and 28·8% (141 of 490) among those age 23–24 years (ptrend=0·0091); prevalence was 31·7% (1057 of 3337) among those age 25–34 years and 22·8% (451 of 1979) among those age 55 and older (ptrend<0·0001). HPV16 prevalence in HIV-negative MSM was 6·7% (15 of 223) among those age 15–18 and 13·9% (166 of 1192) among those age 23–24 years (ptrend=0·0076); the prevalence plateaued thereafter (ptrend=0·72). Similar age-specific patterns were observed for HR-HPV. No significant differences for HPV16 or HR-HPV were found by age for either HIV-positive or HIV-negative MSW. HSIL+ detection ranged from 7·5% (12 of 160) to 54·5% (61 of 112) in HIV-positive MSM; after adjustment for heterogeneity, HIV was a significant predictor of HSIL+ (aPR 1·54, 95% CI 1·36–1·73), HPV16-positive HSIL+ (1·66, 1·36–2·03), and HSIL+ in HPV16-positive MSM (1·19, 1·04–1·37). Among HPV16-positive MSM, HSIL+ prevalence increased with age.InterpretationHigh anal HPV prevalence among young HIV-positive and HIV-negative MSM highlights the benefits of gender-neutral HPV vaccination before sexual activity over catch-up vaccination. HIV-positive MSM are a priority for anal cancer screening research and initiatives targeting HPV16-positive HSIL+.FundingInternational Agency for Research on Cancer.

P433 Sex-based differences in response to tumor necrosis factor inhibitor induction therapy for ulcerative colitis: a pooled analysis of individual patient-level clinical trials data

Abstract Background There are increasing data on sex-based differences in IBD epidemiology, phenotype and outcomes. The purpose of this study was to characterize sex-based differences in response to tumor necrosis factor inhibitor (TNFi) therapies in patients with ulcerative colitis (UC). Methods We conducted a pooled analysis of individual-level data from randomized clinical trials (RCTs) on induction therapy with infliximab and golimumab, accessed through the Yale University Open Data Access platform. We analyzed patients in the treatment and placebo arms separately. Using multivariable logistic regression modeling, we compared the primary outcome of clinical remission, and secondary outcomes of clinical response and mucosal healing, according to clinical trial definitions, between male and female patients with UC at week 6 (golimumab) or week 8 (infliximab) of induction therapy. Effect estimates were expressed as adjusted odds ratios (aOR) and 95% confidence intervals (CI). Results We included 1639 patients (696, 42.5% were women, Table) from two trials using infliximab (ACT-1 and 2) and one trial using golimumab (PURSUIT). Male patients were older than female patients in the treatment arm of the overall cohort [mean age (standard deviation) 31.31 (16.53) years and 29.34 (15.19) years, respectively). All other baseline characteristics, such as sex, race, Mayo score at screening, and concomitant systemic corticosteroid and immunomodulator use were comparable. Compared to female patients, male patients were significantly less likely to achieve clinical remission in both treatment and placebo arms (aOR 0.55, 95% CI 0.31, 0.97 and 0.34, 95% CI 0.15, 0.82, respectively, Figure). Compared to female patients, male patients were significantly less likely to achieve mucosal healing and clinical response in the treatment arm (aOR 0.47, 95% CI 0.27, 0.83 and 0.51, 95% CI 0.29, 0.90, respectively), but there was no significant difference in these secondary outcomes by sex in the placebo arm. Conclusion Based on this analysis of pooled data from RCTs, we demonstrate that clinical remission, response and mucosal healing with TNFi induction therapies are less likely in male patients compared to female patients with UC. Clinical remission is also less likely in male patients on placebo. These finding may lead to further research on sex-based differences in treatment outcomes, including mechanistic studies.

DOP58 Non-white race is associated with decreased efficacy of tumor necrosis factor antagonist therapy in Ulcerative Colitis: A post-hoc analysis of individual level data from golimumab clinical trials

Abstract Background Observational studies suggest non-white patients with inflammatory bowel disease (IBD) have worse clinical outcomes.1 There are limited data on whether race impacts response to biologic therapy. We therefore aimed to evaluate the efficacy of the tumor necrosis factor (TNF) antagonist golimumab comparing white to non-white participants, using individual participant level data from phase 2/3 randomized clinical trials of TNF antagonist therapy in ulcerative colitis (UC). Methods We conducted a pooled analysis of individual-level data from the induction and maintenance trials of golimumab in UC accessible through Yale University Open Data Access Project (YODA). There were insufficient non-white participants in infliximab studies accessible through YODA, precluding meaningful analysis. We analyzed patients in the placebo and treatment arms separately. Our primary outcome was clinical response and secondary outcomes were clinical remission and endoscopic healing according to clinical trial definitions. We compared white and non-white (defined as Black, Asian, or Other race) participants using multivariable logistic regression a priori adjusting for age, sex, treatment group, baseline Mayo score, immunomodulator and corticosteroid use. Effect estimates were expressed as adjusted odds ratios (aOR) and 95% confidence intervals (95% CI). Results A total of 1,006 participants were included in the induction trial (PURSUIT-SC; 18% non-white) and 783 participants in the maintenance trial (PURSUIT-M; 17% non-white). Non-white participants had significantly lower odds of week 6 clinical response (aOR 0.43, 95%CI 0.28–0.66), clinical remission (aOR 0.41, 95%CI 0.22–0.77) and endoscopic remission (aOR 0.48, 95%CI 0.30–0.74) compared to white participants (Figure). Non-white participants also had a lower adjusted odds of week 30 clinical response (aOR 0.64, 95%CI 0.40–1.01), clinical remission (aOR 0.45, 95%CI 0.28–0.74), and endoscopic remission (aOR 0.62, 95%CI 0.41–0.96). By week 54, the lower odds of these outcomes among non-whites were no longer statistically significant (Figure). Conclusion Non-white UC patients were less likely to achieve clinical response, clinical remission, and endoscopic healing with golimumab compared to white patients in these clinical trials. Further studies are needed to understand these differences and whether they are observed with other drugs or outside the context of clinical trials. Reference

Survival Varies By Age and Histology in Classical Hodgkin Lymphoma: A Report from the Children's Oncology Group

Introduction While 5-year event-free (EFS) and overall survival (OS) in Hodgkin lymphoma (HL) generally exceed 85% and 95%, respectively, outcomes may not be as favorable in adolescents and young adults (15 - 39 years [y]) compared to children. Small clinical trials have reported better outcomes for pediatric but not adult patients with mixed cellularity (MC) vs. nodular sclerosing (NS) histology, suggesting the possibility of biologic differences across the age-spectrum in HL. We examined survival by age and histology in patients receiving risk-based, response-adapted therapy for de novo HL on contemporary Children's Oncology Group (COG) trials. Methods This was a pooled analysis of individual-level data from 1,907 patients enrolled on three Phase 3 COG clinical trials for treatment of low-risk (AHOD0431), intermediate risk (AHOD0031) and high-risk (AHOD0831) HL between 2002 and 2012. Histologic subgroups included MC, NS and classical HL, not-otherwise-specified (cHL, NOS). Five-year cumulative incidence of relapse, EFS and OS were compared by age group (&amp;lt;15 y vs. ≥15 y) in the pooled cohort, and in histologic subgroups (MC and non-MC) using the Kaplan-Meier method. Effect modification was confirmed between age and histology. Cox proportional hazards regression models were used to examine the influence of age on EFS and OS, adjusted for race/ethnicity, Ann Arbor stage, B symptoms, bulky disease, receipt of radiation therapy (RT), and the interaction between age and histology; COG study was also included in the model, given that the criteria for response adaptation differed across the trials. Results Between 2002 and 2012, N= 2155 patients 1 - 21 y enrolled on three COG trials, 1,907 (88%) of whom were included in this analysis. Mean age of the cohort was 14.6 y (± 3.5) with N= 871 (46%) &amp;lt;15 y and N= 1,036 (54%) ≥15 y. In total, N= 1,547 patients (81%) had NS histology, N= 108 (6%) had cHL, NOS, and N= 196 (10%) had MC histology; by age, MC histology was present in N= 66 patients (7%) ≥15 y and N= 130 patients (15%) &amp;lt;15 y (p&amp;lt; 0.01). A significantly higher proportion of those ≥15 y vs. younger had B-symptoms at diagnosis (29% vs. 21%, p&amp;lt; 0.01), however the presence of bulky disease did not differ by age. Finally, patients ≥15 y (vs. &amp;lt;15 y) were significantly more likely to receive RT as part of their treatment (72% vs. 63%, p&amp;lt;0.01). Survival: Median follow up was 6.9 years. In unadjusted analyses, 5-year EFS and OS were 83% and 97%, respectively. The 5-year EFS was lower for patients ≥15 y vs. &amp;lt;15 y (85% vs 89%, p&amp;lt;0.01), as was the 5-year OS (96% vs. 99%, p&amp;lt; 0.01). In multivariable models, age ≥15 y (vs. younger) was associated with a 1.4-fold increased risk of EFS (HR: 1.4, 95% CI: 1.1, 1.8, p&amp;lt; 0.01), and more than a 3-fold increased risk of death (OS: HR: 3.1, 95% CI: 1.5, 6.4, p&amp;lt; 0.01). The effect of age on EFS varied by histologic subgroup. Among those with non-MC histology, cumulative incidence of relapse did not significantly differ by age in unadjusted models (Figure A), however 5-year EFS was significantly worse in the older group (Figure B). In multivariable analyses, age ≥15 y (vs. younger) was associated with a 1.3-fold increased risk of EFS (HR: 1.3; 1.03 - 1.7, p= 0.03) (Table). Among patients with MC histology, age ≥15 y (vs. younger) was associated with significantly higher relapse rate (22% vs. 5%, p&amp;lt; 0.01) (Figure C) and significantly worse 5-year EFS (75% vs. 94%, p&amp;lt; 0.01) (Figure D). This remained significant in multivariable models: patients with MC histology who were ≥15 y (vs. younger) had a 3.7-fold increased risk of EFS (HR: 3.7, 95% CI: 1.6, 8.9, p&amp;lt; 0.01) (Table). Conclusion In patients receiving response-adapted therapy for de novo HL on contemporary COG trials, adolescents ≥15 y had worse EFS and OS compared to younger groups. The magnitude of the effect of age was higher in patients with MC disease. Although recent pediatric trials in HL have indicated better survival for some children with MC histology, alternative approaches or novel therapies should be considered for older adolescents with MC disease, whose outcomes appear more like adults. Disclosures No relevant conflicts of interest to declare.

Privacy-protecting estimation of adjusted risk ratios using modified Poisson regression in multi-center studies

BackgroundMulti-center studies can generate robust and generalizable evidence, but privacy considerations and legal restrictions often make it challenging or impossible to pool individual-level data across data-contributing sites. With binary outcomes, privacy-protecting distributed algorithms to conduct logistic regression analyses have been developed. However, the risk ratio often provides a more transparent interpretation of the exposure-outcome association than the odds ratio. Modified Poisson regression has been proposed to directly estimate adjusted risk ratios and produce confidence intervals with the correct nominal coverage when individual-level data are available. There are currently no distributed regression algorithms to estimate adjusted risk ratios while avoiding pooling of individual-level data in multi-center studies.MethodsBy leveraging the Newton-Raphson procedure, we adapted the modified Poisson regression method to estimate multivariable-adjusted risk ratios using only summary-level information in multi-center studies. We developed and tested the proposed method using both simulated and real-world data examples. We compared its results with the results from the corresponding pooled individual-level data analysis.ResultsOur proposed method produced the same adjusted risk ratio estimates and standard errors as the corresponding pooled individual-level data analysis without pooling individual-level data across data-contributing sites.ConclusionsWe developed and validated a distributed modified Poisson regression algorithm for valid and privacy-protecting estimation of adjusted risk ratios and confidence intervals in multi-center studies. This method allows computation of a more interpretable measure of association for binary outcomes, along with valid construction of confidence intervals, without sharing of individual-level data.

Inverse probability weighted Cox model in multi-site studies without sharing individual-level data.

The inverse probability weighted Cox proportional hazards model can be used to estimate the marginal hazard ratio. In multi-site studies, it may be infeasible to pool individual-level datasets due to privacy and other considerations. We propose three methods for making inference on hazard ratios without the need for pooling individual-level datasets across sites. The first method requires a summary-level eight-column risk-set table to produce the same hazard ratio estimate and robust sandwich variance estimate as those from the corresponding pooled individual-level data analysis (reference analysis). The second and third methods, which are based on two bootstrap re-sampling strategies, require a summary-level four-column risk-set table and bootstrap-based risk-set tables from each site to produce the same hazard ratio and bootstrap variance estimates as those from their reference analyses. All three methods require only one file transfer between the data-contributing sites and the analysis center. We justify these methods theoretically, illustrate their use, and demonstrate their statistical performance using both simulated and real-world data.

Validity of Privacy-Protecting Analytical Methods That Use Only Aggregate-Level Information to Conduct Multivariable-Adjusted Analysis in Distributed Data Networks.

Distributed data networks enable large-scale epidemiologic studies, but protecting privacy while adequately adjusting for a large number of covariates continues to pose methodological challenges. Using 2 empirical examples within a 3-site distributed data network, we tested combinations of 3 aggregate-level data-sharing approaches (risk-set, summary-table, and effect-estimate), 4 confounding adjustment methods (matching, stratification, inverse probability weighting, and matching weighting), and 2 summary scores (propensity score and disease risk score) for binary and time-to-event outcomes. We assessed the performance of combinations of these data-sharing and adjustment methods by comparing their results with results from the corresponding pooled individual-level data analysis (reference analysis). For both types of outcomes, the method combinations examined yielded results identical or comparable to the reference results in most scenarios. Within each data-sharing approach, comparability between aggregate- and individual-level data analysis depended on adjustment method; for example, risk-set data-sharing with matched or stratified analysis of summary scores produced identical results, while weighted analysis showed some discrepancies. Across the adjustment methods examined, risk-set data-sharing generally performed better, while summary-table and effect-estimate data-sharing more often produced discrepancies in settings with rare outcomes and small sample sizes. Valid multivariable-adjusted analysis can be performed in distributed data networks without sharing of individual-level data.

Combining distributed regression and propensity scores: a doubly privacy-protecting analytic method for multicenter research

PurposeSharing of detailed individual-level data continues to pose challenges in multi-center studies. This issue can be addressed in part by using analytic methods that require only summary-level information to perform the desired multivariable-adjusted analysis. We examined the feasibility and empirical validity of 1) conducting multivariable-adjusted distributed linear regression and 2) combining distributed linear regression with propensity scores, in a large distributed data network.Patients and methodsWe compared percent total weight loss 1-year postsurgery between Roux-en-Y gastric bypass and sleeve gastrectomy procedure among 43,110 patients from 36 health systems in the National Patient-Centered Clinical Research Network. We adjusted for baseline demographic and clinical variables as individual covariates, deciles of propensity scores, or both, in three separate outcome regression models. We used distributed linear regression, a method that requires only summary-level information (specifically, sums of squares and cross products matrix) from sites, to fit the three ordinary least squares linear regression models. A comparison set of analyses that used pooled deidentified individual-level data from sites served as the reference.ResultsDistributed linear regression produced results identical to those from the corresponding pooled individual-level data analysis for all variables in all three models. The maximum numerical difference in the parameter estimate or standard error for all the variables was 3×10−11 across three models.ConclusionDistributed linear regression analysis is a feasible and valid analytic method in multicenter studies for one-time continuous outcomes. Combining distributed regression with propensity scores via modeling offers more privacy protection and analytic flexibility.

Omega-6 fatty acid biomarkers and incident type 2 diabetes: pooled analysis of individual-level data for 39 740 adults from 20 prospective cohort studies

The metabolic effects of omega-6 polyunsaturated fatty acids (PUFAs) remain contentious, and little evidence is available regarding their potential role in primary prevention of type 2 diabetes. We aimed to assess the associations of linoleic acid and arachidonic acid biomarkers with incident type 2 diabetes. We did a pooled analysis of new, harmonised, individual-level analyses for the biomarkers linoleic acid and its metabolite arachidonic acid and incident type 2 diabetes. We analysed data from 20 prospective cohort studies from ten countries (Iceland, the Netherlands, the USA, Taiwan, the UK, Germany, Finland, Australia, Sweden, and France), with biomarkers sampled between 1970 and 2010. Participants included in the analyses were aged 18 years or older and had data available for linoleic acid and arachidonic acid biomarkers at baseline. We excluded participants with type 2 diabetes at baseline. The main outcome was the association between omega-6 PUFA biomarkers and incident type 2 diabetes. We assessed the relative risk of type 2 diabetes prospectively for each cohort and lipid compartment separately using a prespecified analytic plan for exposures, covariates, effect modifiers, and analysis, and the findings were then pooled using inverse-variance weighted meta-analysis. Participants were 39 740 adults, aged (range of cohort means) 49-76 years with a BMI (range of cohort means) of 23·3-28·4 kg/m2, who did not have type 2 diabetes at baseline. During a follow-up of 366 073 person-years, we identified 4347 cases of incident type 2 diabetes. In multivariable-adjusted pooled analyses, higher proportions of linoleic acid biomarkers as percentages of total fatty acid were associated with a lower risk of type 2 diabetes overall (risk ratio [RR] per interquintile range 0·65, 95% CI 0·60-0·72, p<0·0001; I2=53·9%, pheterogeneity=0·002). The associations between linoleic acid biomarkers and type 2 diabetes were generally similar in different lipid compartments, including phospholipids, plasma, cholesterol esters, and adipose tissue. Levels of arachidonic acid biomarker were not significantly associated with type 2 diabetes risk overall (RR per interquintile range 0·96, 95% CI 0·88-1·05; p=0·38; I2=63·0%, pheterogeneity<0·0001). The associations between linoleic acid and arachidonic acid biomarkers and the risk of type 2 diabetes were not significantly modified by any prespecified potential sources of heterogeneity (ie, age, BMI, sex, race, aspirin use, omega-3 PUFA levels, or variants of the FADS gene; all pheterogeneity≥0·13). Findings suggest that linoleic acid has long-term benefits for the prevention of type 2 diabetes and that arachidonic acid is not harmful. Funders are shown in the appendix.

Overweight, obesity, and risk of cardiometabolic multimorbidity: pooled analysis of individual-level data for 120 813 adults from 16 cohort studies from the USA and Europe

SummaryBackgroundAlthough overweight and obesity have been studied in relation to individual cardiometabolic diseases, their association with risk of cardiometabolic multimorbidity is poorly understood. Here we aimed to establish the risk of incident cardiometabolic multimorbidity (ie, at least two from: type 2 diabetes, coronary heart disease, and stroke) in adults who are overweight and obese compared with those who are a healthy weight.MethodsWe pooled individual-participant data for BMI and incident cardiometabolic multimorbidity from 16 prospective cohort studies from the USA and Europe. Participants included in the analyses were 35 years or older and had data available for BMI at baseline and for type 2 diabetes, coronary heart disease, and stroke at baseline and follow-up. We excluded participants with a diagnosis of diabetes, coronary heart disease, or stroke at or before study baseline. According to WHO recommendations, we classified BMI into categories of healthy (20·0–24·9 kg/m2), overweight (25·0–29·9 kg/m2), class I (mild) obesity (30·0–34·9 kg/m2), and class II and III (severe) obesity (≥35·0 kg/m2). We used an inclusive definition of underweight (<20 kg/m2) to achieve sufficient case numbers for analysis. The main outcome was cardiometabolic multimorbidity (ie, developing at least two from: type 2 diabetes, coronary heart disease, and stroke). Incident cardiometabolic multimorbidity was ascertained via resurvey or linkage to electronic medical records (including hospital admissions and death). We analysed data from each cohort separately using logistic regression and then pooled cohort-specific estimates using random-effects meta-analysis.FindingsParticipants were 120 813 adults (mean age 51·4 years, range 35–103; 71 445 women) who did not have diabetes, coronary heart disease, or stroke at study baseline (1973–2012). During a mean follow-up of 10·7 years (1995–2014), we identified 1627 cases of multimorbidity. After adjustment for sociodemographic and lifestyle factors, compared with individuals with a healthy weight, the risk of developing cardiometabolic multimorbidity in overweight individuals was twice as high (odds ratio [OR] 2·0, 95% CI 1·7–2·4; p<0·0001), almost five times higher for individuals with class I obesity (4·5, 3·5–5·8; p<0·0001), and almost 15 times higher for individuals with classes II and III obesity combined (14·5, 10·1–21·0; p<0·0001). This association was noted in men and women, young and old, and white and non-white participants, and was not dependent on the method of exposure assessment or outcome ascertainment. In analyses of different combinations of cardiometabolic conditions, odds ratios associated with classes II and III obesity were 2·2 (95% CI 1·9–2·6) for vascular disease only (coronary heart disease or stroke), 12·0 (8·1–17·9) for vascular disease followed by diabetes, 18·6 (16·6–20·9) for diabetes only, and 29·8 (21·7–40·8) for diabetes followed by vascular disease.InterpretationThe risk of cardiometabolic multimorbidity increases as BMI increases; from double in overweight people to more than ten times in severely obese people compared with individuals with a healthy BMI. Our findings highlight the need for clinicians to actively screen for diabetes in overweight and obese patients with vascular disease, and pay increased attention to prevention of vascular disease in obese individuals with diabetes.FundingNordForsk, Medical Research Council, Cancer Research UK, Finnish Work Environment Fund, and Academy of Finland.

Diabetes diagnosis and care in sub-Saharan Africa: pooled analysis of individual data from 12 countries

Despite widespread recognition that the burden of diabetes is rapidly growing in many countries in sub-Saharan Africa, nationally representative estimates of unmet need for diabetes diagnosis and care are in short supply for the region. We use national population-based survey data to quantify diabetes prevalence and met and unmet need for diabetes diagnosis and care in 12 countries in sub-Saharan Africa. We further estimate demographic and economic gradients of met need for diabetes diagnosis and care. We did a pooled analysis of individual-level data from nationally representative population-based surveys that met the following inclusion criteria: the data were collected during 2005-15; the data were made available at the individual level; a biomarker for diabetes was available in the dataset; and the dataset included information on use of core health services for diabetes diagnosis and care. We first quantified the population in need of diabetes diagnosis and care by estimating the prevalence of diabetes across the surveys; we also quantified the prevalence of overweight and obesity, as a major risk factor for diabetes and an indicator of need for diabetes screening. Second, we determined the level of met need for diabetes diagnosis, preventive counselling, and treatment in both the diabetic and the overweight and obese population. Finally, we did survey fixed-effects regressions to establish the demographic and economic gradients of met need for diabetes diagnosis, counselling, and treatment. We pooled data from 12 nationally representative population-based surveys in sub-Saharan Africa, representing 38 311 individuals with a biomarker measurement for diabetes. Across the surveys, the median prevalence of diabetes was 5% (range 2-14) and the median prevalence of overweight or obesity was 27% (range 16-68). We estimated seven measures of met need for diabetes-related care across the 12 surveys: (1) percentage of the overweight or obese population who received a blood glucose measurement (median 22% [IQR 11-37]); and percentage of the diabetic population who reported that they (2) had ever received a blood glucose measurement (median 36% [IQR 27-63]); (3) had ever been told that they had diabetes (median 27% [IQR 22-51]); (4) had ever been counselled to lose weight (median 15% [IQR 13-23]); (5) had ever been counselled to exercise (median 15% [IQR 11-30]); (6) were using oral diabetes drugs (median 25% [IQR 18-42]); and (7) were using insulin (median 11% [IQR 6-13]). Compared with those aged 15-39 years, the adjusted odds of met need for diabetes diagnosis (measures 1-3) were 2·22 to 3·53 (40-54 years) and 3·82 to 5·01 (≥55 years) times higher. The adjusted odds of met need for diabetes diagnosis also increased consistently with educational attainment and were between 3·07 and 4·56 higher for the group with 8 years or more of education than for the group with less than 1 year of education. Finally, need for diabetes care was significantly more likely to be met (measures 4-7) in the oldest age and highest educational groups. Diabetes has already reached high levels of prevalence in several countries in sub-Saharan Africa. Large proportions of need for diabetes diagnosis and care in the region remain unmet, but the patterns of unmet need vary widely across the countries in our sample. Novel health policies and programmes are urgently needed to increase awareness of diabetes and to expand coverage of preventive counselling, diagnosis, and linkage to diabetes care. Because the probability of met need for diabetes diagnosis and care consistently increases with age and educational attainment, policy makers should pay particular attention to improved access to diabetes services for young adults and people with low educational attainment. None.

Circulation : Clinical Summaries

<i>Circulation</i> : Clinical Summaries

Abstract B18: Prediagnostic physical inactivity and epithelial ovarian cancer risk and mortality: Evidence from the Ovarian Cancer Association Consortium.

Abstract Despite a growing number of studies evaluating associations between physical activity and epithelial ovarian cancer risk and mortality, the extant evidence is inconclusive, both for epithelial ovarian cancer (EOC) overall, and by individual histotypes. The inconsistency of epidemiological findings may be due to some apparent limitations in the literature. First, individual studies of EOC often have relatively small numbers of case subjects, especially for less common histotypes. Second, physical activity is a complex behavior involving several characteristics that have been measured differently by the various self-report questionnaires utilized among epidemiological studies. Third, many studies in the existing literature have utilized referent groups combining women who report no regular physical activity with women who report some physical activity, which could obscure the relationship between physical activity and EOC risk or mortality, if one exists. Objective: We conducted a pooled analysis of individual-level data from 12 studies from the Ovarian Cancer Association Consortium to investigate the association between recreational physical inactivity and EOC risk and mortality. Methods: In accordance with the 2008 Physical Activity Guidelines for Americans, women were classified as inactive, our exposure of interest, if they reported participating in recreational physical activities less than once per week, on average, throughout adulthood. To decrease the likelihood of reverse causation as an explanation for observed associations, studies using activity data from the reference date only were excluded in sensitivity analyses. For risk assessment, case data were from 9,024 women aged 18 years and older with a primary diagnosis of borderline or invasive EOC including serous, mucinous, endometrioid or clear cell tumors. Control data were from 13,643 women aged 18 years and older who had at least partially intact ovaries and no histories of ovarian cancer. We employed both meta-analytic and combined individual-data approaches to evaluate the associations between physical inactivity and EOC risk. For mortality outcomes, follow-up data and vital status were available for 6,807 cases. Using a combined dataset, multivariable Cox proportional hazards regression analyses were utilized to generate Hazard Ratios (HRs) and 95% CIs for the association between physical inactivity and all-cause mortality. We also examined risks of EOC-specific mortality and disease progression in a smaller subset of patients with available follow-up data. Results: In risk analyses, we observed a significant positive association between recreational physical inactivity and risk of EOC (OR=1.39, 95% CI: 1.18-1.63, p&amp;lt;0.001). Significant risk elevations ranging between 30% to 55% among inactive women were also observed across the individual EOC histotypes. In survival analyses, there were 3,899 all-cause deaths for analysis. We observed a significant increased multivariable adjusted risk of death among inactive women in comparison to more active women (HR=1.14, 95% CI: 1.06-1.23, p&amp;lt;0.001). We also observed increased risk of EOC-specific mortality, but the association did not achieve statistical significance (HR=1.10, 95% CI: 0.95-1.28, p=0.211). Lastly, we observed a significantly higher risk of disease progression among inactive patients (HR=1.22, 95% CI: 1.09-1.37, p=0.001). Conclusions: These data suggest that recreational physical inactivity in adulthood is associated with an increased risk of EOC. Our data also supports an association between inactivity with risk of disease progression and mortality. These associations were robust to adjustment for relevant confounders and to sensitivity analyses designed to decrease the likelihood of reverse causation as an explanation for observed associations. Given the growing prevalence of inactivity in the U.S. and worldwide, these findings are relevant to public health and cancer prevention and control. Citation Format: Rikki A. Cannioto, Michael LaMonte, Ovarian Cancer Association Consortium (OCAC), Kirsten B. Moysich. Prediagnostic physical inactivity and epithelial ovarian cancer risk and mortality: Evidence from the Ovarian Cancer Association Consortium. [abstract]. In: Proceedings of the AACR Special Conference on Advances in Ovarian Cancer Research: Exploiting Vulnerabilities; Oct 17-20, 2015; Orlando, FL. Philadelphia (PA): AACR; Clin Cancer Res 2016;22(2 Suppl):Abstract nr B18.