Selected Conference Proceedings
14th Annual Epidemiology, Biostatistics and Occupational Health Student Research Day

Message from the Epidemiology, Biostatistics and Occupational Health Student Society

Selected abstracts from the 14th Annual Epidemiology, Biostatistics and Occupational Health Student Research Day, hosted by the Epidemiology, Biostatistics and Occupational Health Student Society on March 16, 2018 at the Montreal Neurological Institute. This event celebrates the research done by students in EBOH at McGill University. Special thanks to the abstract reviewers, presentation judges, administrative staff, faculty donors, and organizing committee, as well as the volunteers, without whom this event would not have been possible.

Preventable burden of infection-associated cancers among American adults in 2014

Karena Volesky1, Mariam El-Zein1, Darren R. Brenner2, Christine M. Friedenreich2, Yibing Ruan2, Stephen Walter3, Abbey E. Poirier2, Eduardo L. Franco1 on behalf of the ComPARe Study Group.
1McGill University
2Alberta Health Services
3McMaster University

Background: Infections play an important role in the etiology of cancer at several anatomical sites. Certain cancer-causing infections are avoidable. For instance, vaccination against human papillomavirus (HPV) and hepatitis B (HBV) infections or antibiotic treatment for Helicobacter pylori are proven preventive strategies. We estimated the proportion of incident cancer cases in 2014 attributable to HBV, HPV and H. pylori among American adults aged 20 and older.

Methods: For each of these infections, the prevalence and relative risk estimates required to calculate population attributable risks (PARs) were abstracted from systematic literature searches. PARs were applied to cancer incidence data from the Centers for Disease Control and Prevention and the National Cancer Institute. HBV and H. pylori prevalence estimates were obtained from the National Health and Nutrition Examination Survey for the years 2003-2004 and 1999-2000, respectively. Measurement error in earlier H. pylori prevalence and relative risk estimates was corrected for the gold standard assay.

Results: The estimated number of cancer cases attributable to HPV was 9,300 among men and 23,000 among women. For H. pylori, it was 5,700 cases among men and 5,200 among women, whereas for HBV it was 2,100 cases among men and 400 among women. Of all cancers diagnosed among Americans in 2014, 2.86% (95%CI: 2.18-3.56%) were attributable to these three infectious agents, 2.13% (95%CI: 1.36-2.89) in men and 3.58% (95%CI: 3.00-4.23) in women.

Conclusion: If HPV, H. pylori, and HBV were eliminated with existing interventions, an estimated 45,600 of the ~1.5 million cancer cases diagnosed among Americans in 2014 would have been avoided. These findings highlight the importance of increasing HPV and HBV vaccination coverage and treating H. pylori.

Karena Volesky is a PhD Student (Epidemiology) at McGill University. She can be reached at

A data-adaptive procedure for the optimal discretization of the timeline for longitudinal causal inference methods

Steve Ferreira Guerra1, Amélie Forget2 , Lucie Blais1,2, Mireille E. Schnitzer1
1Faculté de Pharmacie, Université de Montréal
2Research Center, Hôpital du Sacré-Coeur de Montréal

Background: In pharmacoepidemiology, administrative databases have become abundantly used to conduct research on drug safety and effectiveness. In longitudinal settings, administrative databases provide information on data that is collected in real-time. In such settings, a potential source of bias is time-dependent confounding, which can be addressed with longitudinal causal inference methods such as Marginal Structural Models (MSM). However, these methods usually rely on a discretization of the patient timeline that may not reflect the underlying continuous nature of administrative data. Thus bias may result when the discretization is arbitrarily chosen by the researcher, which is common practice.

Methods: We develop a new method for the automatic selection of an optimal data timeline discretization for use with Longitudinal Targeted Maximum Likelihood Estimation (LTMLE) of MSMs. We use a simulation study to evaluate the bias-variance trade-off of such a method.

Results: We show how coarsening changes the underlying true population of interest as well as how it creates bias of our parameter of interest and affects it's variance. We also demonstrate the performance of our selection procedure in simulated settings.

Conclusion: In conclusion, this research is of utmost importance since it reflects flaws in common practices and addresses this problem by evaluating the impact of discretization and by proposing an automatic optimal discretization selection method that seems to perform appropriately.

Steve Ferreira Guerra is a PhD Student (Biostatistics) at McGill University. He can be reached at

Cardiovascular-Related Morbidity and Mortality in Women with a History of Pregnancy Complications: A Systematic Review and Meta-Analysis

Sonia M. Grandi1,2, Kristian B. Filion1,2,3, Sarah Yoon1,2, Henok Ayele1,2, Carla Doyle1,2, Jennifer A. Hutcheon4, Graeme N. Smith5, Joel Ray6, Robert W. Platt1,2,7,8
1Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, QC
2Centre for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, Montreal, QC
3Department of Medicine, McGill University, Montreal, QC
4Department of Obstetrics and Gynaecology, Faculty of Medicine, University of British Columbia, Vancouver, BC
5Department of Obstetrics and Gynaecology, School of Medicine, Queen's University, Kingston, ON
6Department of Medicine and the Institute of Health Policy, Management and Evaluation, University of Toronto, Toronto, ON
7McGill University Health Center Research Institute, Montreal, QC
8Department of Pediatrics, McGill University, Montreal, QC

Introduction: Women with a history of pregnancy complications often have elevated cardiovascular disease (CVD) risk factor levels at or shortly after delivery and that these effects may persist long-term. However, clinical guidelines recommend post-partum follow-up only in women with a history of preeclampsia or preterm birth. We therefore performed a systematic review of observational studies to examine the association between pregnancy complications and the risk of subsequent CVD.

Methods: We systematically searched PubMed, MEDLINE and EMBASE (via Ovid), CINAHL, and the Cochrane Library for observational studies investigating the association between pregnancy complications, including hypertensive disorders in pregnancy (HDP), gestational diabetes (GDM), low birth weight, placental abruption, preterm birth, small-for-gestational-age (SGA) at birth, stillbirth, and pregnancy loss, and subsequent CVD. Studies were grouped by pregnancy complication and design to facilitate between study comparisons. Quality assessment was performed using the ROBINS-I tool. Random-effects likelihood ratio meta-analyses were performed for five of the pregnancy complications.

Results: We identified 13,969 potentially relevant publications, of which 84 met our inclusion criteria. Follow-up ranged from 0 to 55 years, and sample sizes varied from 250 to 2,000,000 women. The pooled results for HDP, GDM, placental abruption, preterm birth, and stillbirth suggest that these complications are associated with an ~2-fold increased risk of subsequent CVD (range: OR 1.67 [95% intrinsic confidence interval 1.31-2.12] to 2.75 [95% intrinsic confidence interval 2.33-3.23]). The findings for pregnancy loss, low birth weight, and SGA were heterogeneous across studies but suggested no increased risk of CVD. The included studies were found to be of varying quality, largely due to insufficient adjustment for known confounders.

Conclusion: Women with a history of the included pregnancy complications are at increased risk of subsequent CVD. The findings support the importance of continuous follow-up and risk-factor management in these women beyond the post-partum period.

Sonia Grandi is a PhD Student (Epidemiology) at McGill Univeristy. She can be reached at

Androgen deprivation therapy for prostate cancer and the risk of bladder cancer: a systematic review of observational studies

Christina Santella1,2, Julie Rouette1,2, Michael Brundage3, Kris Filion1,2,4, Laurent Azoulay1,2,4
1Division of Clinical Epidemiology, Jewish General Hospital/McGill University
2Department of Epidemiology, Biostatistics, and Occupational Health, McGill University
3Division of Cancer Care and Epidemiology, Cancer Research Institute Queen’s University
4Faculty of Medicine, McGill University

Background: Recent observational studies have investigated the association between androgen deprivation therapy (ADT) and bladder cancer in patients with prostate cancer. The results of the studies all suggested a highly protective potential of ADT against bladder cancer. We conducted a systematic review of these studies to assess the methodological strengths and limitations of the evidence.

Methods: We systematically searched Medline, EMBASE, Healthstar, Cochrane Library Online, Science Citation Index, Dissertation Abstracts Online,, and WHO, from inception to present to identify all studies exploring the association between ADT and bladder cancer. Studies were included if they included patients with prostate cancer who underwent ADT, reported at least one bladder cancer outcome (incidence or recurrence), and provided at least one effect measure of the association of interest. There were no restrictions on date or language. We assessed overall study quality using the ROBINS-I tool and evaluated the presence of time-related biases, inclusion of a lag period, confounding, duration of follow-up between ADT initiation and bladder cancer, and inclusion of prevalent users.

Results: Our systematic review included three observational studies. Two studies evaluated the association between ADT and bladder cancer recurrence, and one looked at ADT and bladder cancer incidence. All studies reported a decreased risk of bladder cancer with ADT use. Based on ROBINS-I, all three studies had a serious risk of bias. All studies had time-related biases, did not consider a lag period, did not account for important confounding factors, had short follow-up, and two included prevalent users.

Conclusion: The observational studies examining the use of ADT and the risk of bladder cancer in patients with prostate cancer have important methodological limitations. Therefore, the conclusions that can be drawn from these studies are limited. Future studies addressing these limitations are needed to conclusively evaluate this association.

Christina Santella is a MSc Student (Epidemiology) at McGill Univeristy. She can be reached at

Virtual Pooling as a Privacy-preserving Analysis Tool to Estimate Covariate Hazard Ratio (HR) of Cox Proportional Hazard Model

Lamin Juwara1, Alexandra Schmidt1, Paramita Saha-Chaudhuri1
1Department of Epidemiology, Biostatistics and Occupational Health, McGill University

Background: The use of health care data is constrained by legitimate privacy concerns in multi-center (or distributed data) studies, preventing the creation of a single dataset of covariate information of all participants. In this study, we explore the application of specimen pooling as a privacy-preserving tool for estimating hazard ratio (HR) of a covariate for a time to event outcome.

Methods and Results: For a nested case-control subset, by utilizing the equivalence between the Cox proportional hazards (PH) model and conditional logistic model, we estimate the HRs using only the aggregate covariates. The pooled estimates (exposure HR 1.52, SE 0.59; confounder HR -2.04, SE 0.30) are shown to be similar to individual level (unpooled) covariate effect estimates (exposure HR 1.51, SE 0.48; confounder HR -1.99, SE 0.22). Thus, aggregate levels of covariates can be used to estimate HRs of interest without revealing individual participant’s data. As expected, we observe similar effect estimates between the unpooled cox model and pooled conditional logistic models with inflated standard errors in larger pool sizes. Additionally, effect modifiers can be accommodated and consistently estimated. The approach is demonstrated using extensive simulation studies and a real data example.

Conclusion: Aggregate covariates can be used within a nested case-control subcohort to estimate HR of a Cox PH model. Such an approach preserves privacy while providing reasonable estimates of relevant HRs including confounders and effect modifiers.

Lamin Juwara is a MSc Student (Biostatistics) at McGill Univeristy. He can be reached at

People who carry luggage: opportunities for strengthening the community health information system in two informal settlements of Nairobi, Kenya

Ga Eun Lee1,2, Pauline Bakibinga2
1Department of Epidemiology, Biostatistics and Occupational Health, McGill University
2African Population and Health Research Center

Background: In 2008, the Ministry of Health (MOH) in Kenya implemented the Community Health Strategy, an approach aimed to effectively involve and empower communities to take ownership of improving their own health. A key component of the Strategy is the Community Health Information System (CHIS) which refers to the information generated, collected, analyzed, and used by the community and at levels beyond for planning, monitoring, and decision-making at the community level. However, literature shows that the current performance of Kenya’s CHIS is unsatisfactory and highly variable across communities.

The Performance of Routine Information System Management (PRISM) framework states that routine information systems are affected by internal determinants including technical, environmental/organizational, and behavioural determinants that affect health system performance and consequently lead to better health outcomes. Ahead of implementing a CHIS strengthening project, a baseline qualitative assessment was conducted. The purpose of the qualitative study was to use the PRISM framework to explore the experiences of women of reproductive age (WoRA), community health volunteers (CHVs), and health providers using the current CHIS and identify opportunities for strengthening the system in Embakasi and Kamukunji sub-counties of Nairobi County, Kenya.

Methods: Purposive sampling was used whereby CHVs identified pregnant women and mothers with children less than one year old. Health providers in selected health facilities and members of the Sub-County Health Management Teams were also purposively selected. In total, 5 FGDs with CHVs, 13 FGDs with WoRA, and 17 key-informant interviews with health providers were conducted. All interviews were audio recorded, translated into English, and transcribed into word-processing files that were uploaded to NVivo 11 for coding and analyzed using a thematic framework approach.

Results: CHVs use data collection tools other than MOH tools due to disadvantages of the current system including cumbersome paper-based tools, lack of resources, and lengthy time taken to collect data. Health providers promote a culture of information use, but structures for disseminating and using community health information were rarely discussed by WoRA and CHVs. Both CHVs and WoRA do not appear understand the purpose of data collection and utilization. This lack of understanding may contribute to the community’s prevailing negative attitudes towards data collection activities by CHVs. Mobile phones were discussed as a method of improving the current system.

Conclusion: The findings reveal the necessities in supporting communities in the collection and utilization of community health information and in making data relevant to those on the ground.

Ga Eun Lee is a MSc Student (Public Health) at McGill Univeristy. She can be reached at

Selection of controls in the EnvIMS case-control study: A tribute to Sholom Wacholder

Sean MacKnight1, Jacob Mosseri2, and Christina Wolfson1
1Department of Epidemiology, Biostatistics and Occupational Health, McGill University
2Faculty of Applied Science and Engineering, University of Toronto

Background: Case-control studies are the most frequently conducted etiological study in multiple sclerosis (MS). In theory, the control series should be representative of the source population from which the cases arose (Wacholder 1992) but researchers rarely examine their representativeness.

Methods: The Environmental Risk Factors in MS (EnvIMS) Study is a classical case-control study incorporating population-based sampling in four countries: Canada, Italy, Norway, and Sweden. Controls were recruited using random digit dialing (RDD) in Canada and selected from population registries in Italy, Norway, and Sweden. We compared weight, height, smoking status, and education in the EnvIMS control series with national reference values, taken from censuses and population surveys, to assess the “representativeness” of controls.

Results: Of the compared variables, controls were found to have higher levels of education than the general population in all 4 countries. In comparison with the general population, Canada’s controls were similar in the distributions of weight and height, but male controls were less likely to be smokers. The control series from Italy were similar to the general population in weight and smoking distributions but tended to be shorter. In Norway, the distribution of weight, height, and smoking in the controls were all similar to the general population. Finally, the Swedish controls were similar to the general population with respect to weight and height, but were less likely to be smokers.

Conclusion: In general, we find that the EnvIMS control series adequately represents the study populations. The higher education in study participants is a known phenomenon in observational studies. How these differences may impact the study findings for specific risk factors is currently under study. There did not seem to be any consistent difference between controls selected through population registries and those selected through RDD.

Wacholder S, et al. Selection of controls in case-control studies. 1. Principles. AJE, 1992; 135: 1019–28.

Sean MacKnight is a MSc Student (Public Health) at McGill Univeristy. He can be reached at

Use of dipeptidyl peptidase-4 inhibitors and new-onset rheumatoid arthritis in patients with type 2 diabetes

Antonios Douros1,2,3, Devin Abrahami1,2, Hui Yin1, Oriana Hoi Yun Yu1,4, Christel Renoux1,2,5, Marie Hudson1,6, Laurent Azoulay1,2,7
1Centre for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, Montreal, Canada
2Department of Epidemiology, Biostatistics, and Occupational Health, McGill University, Montreal, Canada
3Charité - Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Institute of Clinical Pharmacology and Toxicology, Berlin, Germany
4Division of Endocrinology, Jewish General Hospital, Montreal, Canada
5Department of Neurology and Neurosurgery, McGill University, Montreal, Canada
6Division of Rheumatology, Jewish General Hospital, Montreal, Canada and Department of Medicine, McGill University, Montréal, Canada
7Gerald Bronfman Department of Oncology, McGill University, Montreal, Canada

Background: Case reports have suggested a link between dipeptidyl peptidase-4 (DPP-4) inhibitors, antidiabetic drugs used as second-to-third-line treatments, and the incidence of rheumatoid arthritis (RA). Since the DPP-4 enzyme is involved in several immunologic processes and possibly in RA pathophysiology, further research is warranted. Thus, this population-based study aimed to determine whether use of DPP-4 inhibitors is associated with the incidence of RA.

Methods and Results: Using the United Kingdom Clinical Practice Research Datalink, we conducted a cohort study among 144,603 patients with type 2 diabetes initiating antidiabetic drugs between 2007 and 2016. Hazard ratios (HRs) with 95% confidence intervals (CIs) of incident RA were estimated using time-dependent Cox proportional hazards models, comparing use of DPP-4 inhibitors with use of other antidiabetic drugs. A 6-month exposure lag period was imposed for latency and diagnostic delays. Secondary analyses assessed whether there was a duration-response relation or a drug-specific effect for different DPP-4 inhibitors (sitagliptin, saxagliptin, linagliptin). During 567,169 person-years of follow-up, 464 patients were newly-diagnosed with RA (crude incidence rate: 81.8 per 100,000/year). Compared with use of other antidiabetic drugs, use of DPP-4 inhibitors was not associated with an increased risk of RA (82.3 versus 79.3 per 100,000/year; HR: 0.99, 95% CI: 0.75, 1.29), with no evidence of duration-response relation or drug-specific effects.

Conclusion: In this large population-based study, use of DPP-4 inhibitors was not associated with an increased risk of incident RA. These findings provide reassurance regarding the safety of DPP-4 inhibitors on the incidence of this autoimmune condition.

Antonios Douros is a Post-Doctoral Fellow (Epidemiology)) at McGill Univeristy. He can be reached at

Trends in the Pharmacological Treatment of Benign Prostatic Hyperplasia in the United Kingdom from 1998 to 2016: A Population-Based Cohort Study

Henok Tadesse Ayele1,2, Pauline Reynier2, Laurent Azoulay1,2,3, Robert W. Platt1,2, Serge Benayoun4, Olli Saarela5, Kristian B. Filion1,2,6
1Department of Epidemiology, Biostatistics, Occupational Health, McGill University Montréal, Canada
2Center for Clinical Epidemiology, Lady Davis Institute, Jewish General Hospital, Montréal, Canada
3Gerald Bronfman Department of Oncology, McGill University, Montréal, Canada
4Department of Surgery, Urology Unit, Hôpital Maisonneuve-Rosemont, Montréal, Canada
5Dalla Lana School of Public Health, University of Toronto, Toronto, Canada
6Department of Medicine, McGill University, Montréal, Canada

Background: Five-alpha reductase inhibitors (5αRIs) and α-blockers are widely prescribed to manage the symptoms of benign prostatic hyperplasia (BPH). However, trends in the use of BPH medications are not well studied. Our objective was to describe patterns of BPH medications use in United Kingdom (UK) from 1998-2016.

Methods: We created a cohort of men aged ≥40 years with newly diagnosed BPH between 1998 and 2016 using the UK Clinical Practice Research Datalink (CPRD), a clinical database containing records from over 700 general practitioner practices. Using Poisson regression, we estimated the annual prescription rate of 5αRIs, α-blockers, and combination therapy (5αRIs and α-blockers). We estimated age-adjusted rate ratios (RRs) comparing the prescription rates in 2016 to those of 1998.

Results: We identified 94,598 patients with newly diagnosed BPH, from 208,230 patients with BPH diagnosis in the CPRD, generating 536,471 person-years (PYs) of follow-up. We excluded patients with the diagnosis occurred before they turn 40 years of age (n=1,248), January 1, 1998 (n=69,265), or up to standard (UTS) history of less than 1 year (n=43,119). Only 29.3% patients received 5αRIs, α-blockers, or combination therapy at cohort entry. Overall, the rate of prescription increased from 238.7 (95% CI: 230.9-246.7) per 100 PYs in 1998 to 414.2 (95% CI: 409.6-419.0) per 100 PYs in 2016, representing a 50% (RR: 1.5, 95% CI: 1.5-1.6) increase. The α-blockers accounted for over 68% of all new prescriptions. Between 1998 and 2016, the prescription rate of 5αRIs and α-blockers increased by 10% (RR: 1.1, 95% CI: 0.97-1.13) and 30% (RR: 1.3, 95% CI: 1.2-1.3), respectively, whereas prescription of combination therapy increased 24 times (RR: 24.3, 95% CI: 18.0-32.7).

Conclusion: The rate of BPH medication use increased substantially between 1998 and 2016 in the UK, with the greatest relative increase observed with combination therapy.

Henok Tadesse Ayele is a Post-Doctoral Fellow (Epidemiology) at McGill Univeristy. He can be reached at

Efficacy of a Carrageenan gel Against Transmission of Cervical HPV (CATCH): interim analysis of a randomized, double-blind, placebo-controlled trial

Sindy Magnan1, Joseph E. Tota2, Mariam El-Zein1, Ann N Burchell3, John T Schiller4, Alex Ferenczy5, Pierre-Paul Tellier6, Francois Coutlee7, Eduardo L. Franco1
1McGill University, Division of Cancer Epidemiology, Montreal, Canada
2National Cancer Institute, Division of Cancer Epidemiology and Genetics, Rockville, USA
3Li Ka Shing Knowledge Institute, Department of Family and Community Medicine, Toronto, Canada
4National Cancer Institute, Department of Health and Human Services, Bethesda, USA
5McGill University, Department of Pathology, Montreal, Canada
6McGill University, Department of Family Medicine, Montreal, Canada
7Centre Hospitalier de l’Université de Montréal, Département de Microbiologie et infectiologie, Montréal, Canada

Background: Carrageenan has been identified as a potent human papillomavirus (HPV) infection inhibitor in preclinical studies. Our objective was to evaluate the efficacy of a carrageenan-based gel in reducing the risk of genital HPV infections among sexually active women.

Methods: Between January 2013 and June 2017, 280 women aged 18 years and older were randomly assigned to a carrageenan or a placebo gel to be self-applied every other day for the first month and prior to and following each intercourse during the one-year study period. Sociodemographic, behavioural and sexual history data were collected using computer-assisted self-administered questionnaires. We used Roche’s Linear Array assay to detect and genotype 36 HPV types in vaginal samples. HPV types were categorized based on tissue tropism and oncogenicity: alphapapillomavirus subgenus 1 (low oncogenic risk), 2 (high oncogenic risk) and 3 (commensal). The primary outcome was incidence of a newly detected infection by an HPV type that was not present at baseline. We computed hazard ratios (HR) and 95% confidence intervals (CI) using Cox models and mixed effects survival-time models.

Results: Baseline and follow-up characteristics were well balanced between arms. The median follow-up time was 9.2 months (IQR: 2.6-13.2). 59 participants in the carrageenan arm and 78 participants in the placebo arm got infected by at least one new HPV type (HR: 0.64; 95% CI: 0.45-0.89). A lower incidence was consistently observed for all alphapapillomavirus subgenera (HR range: 0.47-0.69). When considering all new HPV types acquired by each participant (not only the first infection), 139 infections occurred in the carrageenan arm versus 198, in the placebo arm (HR: 0.66; 95% CI: 0.46-0.93; taking into account the correlated data structure).

Conclusion: Our trial’s interim analysis suggests that using a carrageenan-based gel can reduce the risk of new genital HPV infections of both low and high oncogenic risk in women.

Sindy Magnan is a PhD Student (Epidemiology) at McGill Univeristy. She can be reached at