Study setting

Canada is a low TB-incidence country, with the majority of TB cases (70%) diagnosed in persons born outside of Canada, many from high TB-incidence countries.12 BC is a low TB-incidence region with an annual TB incidence of 6.0 per 100 000 people.12

Notably, over 20% of Canada's population is foreign-born, one of the highest among the G20 economies.13 BC is a Canadian province with a population of ∼4.4 million people and is one of the major immigrant-receiving provinces in Canada. In 2011, 28% of BC's population was estimated to be born outside of Canada.14 Immigration patterns to BC have changed over the past 5 decades, with an increase in migrants coming from high TB-incidence countries. In 1965, the major source countries for immigration to BC were Great Britain (34%) and the USA (17%), followed by Italy (6%), Germany (6%) and China (6%).15 In contrast, between the years 2000 and 2015, approximately half of all immigrants to BC came from three high TB-incidence source countries—China (23%), India (15%) and the Philippines (12%).16

In BC, the BC Centre for Disease Control (BCCDC) is the centralised public health agency that is responsible for diagnosis and treatment of the majority of active TB and LTBI patients in the province.17 The BCCDC also maintains the Provincial TB Registry which includes diagnostic and treatment data on active TB cases. Mandatory notification by public health partners and routine reporting from the centralised provincial mycobacteriology laboratory and provincial pharmacy makes the Provincial TB Registry virtually complete for active TB.

Data sources and linkages

This study will take advantage of a recently established permanent data linkage between Immigration, Refugees and Citizenship Canada (IRCC) and Population Data BC. IRCC is the Canadian Federal Government agency that grants citizenship, facilitates the arrival of immigrants, provides protection to refugees and offers assistance for migrants to Canada.18 IRCC data on Permanent Residents have been provided to Population Data BC by the IRCC Research and Evaluation Branch for the purpose of facilitating immigrant health-related research projects. De-identified data pertaining to citizenship and immigration, such as immigration class, year of arrival and country of origin, are available to researchers by request.19

Population Data BC is a multiuniversity data resource housed at the University of BC. Population Data BC supports research access to individual-level, de-identified longitudinal data on BC residents. Population Data BC holdings represent one of the world's largest collections of administrative data, including healthcare, health services and population health data.20 These data sets are all linkable to each other and to additional external data sets, where approved by the data provider.

The IRCC database has been linked previously with Population Data BC using a minimum 6 of a 7-point matching algorithm (First Name, Middle Name, Surname, Sex, Year of Birth, Month of Birth, Day of Birth) (Personal communication from Population Data BC, 21 October 2015). From this linked data set, we requested data on all foreign-born permanent residents who established residency in BC between 1985 and 2013. The requested data include data on physician billings, hospitalisations, drugs dispensed from community pharmacies, vital statistics, HIV testing, cancer diagnoses, chronic kidney disease (CKD) and dialysis treatment, and all TB and LTBI testing and treatment data in BC. Data will be linked between each data source using unique scrambled patient identification numbers (table 1; figure 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Schematic of database linkages.

Study design and study population

This is a retrospective population-based cohort study. The study population will include all foreign-born permanent residents who have landed in Canada between 1 January 1985 and 31 December 2013, and established residency in BC at any point during this same period.

Individuals will be identified as being resident in BC when they have registered in the provincial Medical Services Plan (MSP).38 We will exclude individuals from our cohort who did not acquire an MSP number. MSP is the Universal Health Insurance Programme administered by the provincial government of BC. It insures medically required services provided by physicians and healthcare practitioners. Enrolment with MSP is mandatory for all eligible BC residents and their dependents. To be eligible, residents must be citizens or permanent residents of Canada and must be physically present in BC at least 6 months in a calendar year. Depending on an individual's income, MSP coverage may be free, or may require monthly premiums up to $C75 per month. Thus, MSP coverage can be considered a good proxy measure for residence in BC.

Measuring follow-up time and primary outcome

The primary outcome measure will be time to active TB diagnosis. TB diagnosis will be identified based on BCCDC TB Registry data, and will include all TB sites (ie, pulmonary and extrapulmonary TB), either microbiologically or clinically confirmed. A TB diagnosis is established in the BCCDC registry based on the Canadian TB Reporting System Guidelines.39

Calculation of follow-up time will begin for all individuals at their index event (figure 2). The index event will be identified as occurring 90 days before their first MSP registration date. We will use the MSP registration as a proxy to ensure that individuals are resident in BC. We will start the collection of follow-up time at 90 days before first MSP registration, to account for the mandatory 90-day waiting period for starting MSP. For example, an individual who moved to BC from overseas and came under MSP coverage on 1 March would be treated in our study as having been a BC resident since January 1st of the same year. Their exposure and time at risk for study outcomes would then be treated as beginning on 1 January. This will allow us to capture TB outcomes occurring within the first 3 months of residency in the province, while an individual is still awaiting registration in MSP. Follow-up will be censored at active TB diagnosis, or after the end of MSP coverage, death or end of the study period (31 December 2013).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Defining study follow-up periods—index date (entry into cohort) and breaks in MSP registration. †End of follow-up due to earliest of: active TB diagnosis, death, end of MSP registration, or end of study period (31 December 2013). BC MSP, British Columbia Medical Services Plan (universal health insurance plan).

Potential risk factors for active TB

We will identify risk factors targeted in the Canadian TB Standards as high-risk conditions for TB development or TB reactivation.7 See online supplementary appendix (tables 1, 2) for detailed definitions of variables.

Demographic and immigration-related factors will be identified primarily from IRCC records. Variables will include sex, age at arrival to Canada (in years), age at TB diagnosis or censoring (in years), years since arrival in Canada, year of arrival in Canada, immigration classification (ie, economic, refugee or other) and TB incidence in country of origin in year of arrival to Canada.

TB-related risk factors will be identified primarily using data from the Provincial TB Registry, supplemented by data extracted from the literature:

For the TB incidence rate in the individual's country of origin, we will use country-level WHO TB incidence data (all forms of TB/100 000 population), as reported to the WHO for each country per year since 1990.40 For years prior to 1990 (1985–1989), we will apply 1990 TB incidence rates. We will investigate the impact of any changes in WHO definitions of confirmed active TB cases occurring over this time period, and account for this (ie, through the use of weights), if necessary.

BCG vaccination will be verified if the individual has a BCG-positive status recorded in the TB Registry. For individuals without records in the Provincial TB Registry, we will use the World BCG Atlas to identify if an individual likely was previously vaccinated with BCG, based on the reported BCG vaccination policies in their birth country and year of birth.41

Using linked BCCDC TB Registry data, we will identify if an individual was a contact of an active TB case in BC, if they presented with an abnormal chest X-ray, and any dates of positive tuberculin skin test (TST) and/or positive Interferon Gamma Release Assay (IGRA) if measured. We will identify if the individual was referred by IRCC for postlanding TB surveillance (ie, IRCC requires immigrants and refugees with chest X-ray or historical evidence of prior TB, when detected in the Prearrival TB Screening Programme, to report to local public health authorities within 30 days of arrival for examination and follow-up).7 ,42

Finally, we will identify if and when an individual initiated and/or completed LTBI treatment. LTBI treatment completion will be defined as being dispensed at least 80% of recommended doses (ie, completion of 210 doses of isoniazid or 76 doses of rifampin, based on recommended guidelines).7

Medical comorbidities considered high-risk for the development of active TB and which have been targeted for LTBI screening by the Canadian TB Standards will be identified.7 These include HIV/AIDS, CKD, cancer, diabetes, medical immunosuppression, solid organ transplant and silicosis (See online supplementary appendices tables 1 and 2). Disease registries will be used as the gold standard for disease diagnoses, where available. We will supplement disease registry data with data from health administrative databases, using validated algorithms where possible. For the purposes of estimating the impact of medical comorbidities on TB risk, we will estimate risk factor exposure start and end dates separately for each disease based on expert opinion and literature review (see online supplementary appendix table 1). For example, for the impact of drugs on TB risk, we will allow a 30-day lag period for a drug to take effect. For chronic diseases such as diabetes, we will assume that the effect on TB risk begins 90 days before the first date where we have confirmation of diabetes diagnosis to allow for a period of diagnostic delay.

As the Provincial Renal Agency data for CKD and solid organ transplants are not available prior to 2003, we will develop MSP physician billing/hospitalisation-based data algorithms and compare these against patient registry data between the years 2003 and 2013. Algorithms based on International Classification of Diseases (ICD) diagnostic codes, and inpatient and outpatient procedure codes, with the highest accuracy for identifying CKD and solid organ transplants will be identified, and their reliability as indicators will be described.

Analysis plan for each objective

Objective 1: To identify individual risk factors, we will examine the relationship between each potential baseline risk factor and the diagnosis of active TB using Kaplan-Meier survival curves and the log-rank test for categorical risk factors and for some numeric risk factors that naturally split into clinically relevant categories. For other numeric risk factors we will use proportional hazards regression, possibly with regression splines.

Objective 2: To create a predictive model for the risk of developing active TB, we will first randomly partition our complete cohort into three analysis sets with a 2:1:1 ratio. These will be referred to as our training set (50%), validation set (25%) and test set (25%). For the primary predictive models, we will use an extended proportional hazards regression model43 which may include time-varying coefficients, time-varying predictors as well as terms for interactions among the predictors. Our candidate predictive models will be developed based on modern modelling strategies (eg, the lasso) using only the training set. We will then compare the optimism-corrected estimates of performance of our candidate models (ie, c-index measuring concordance between predictions and actual survival) in the validation set. We will choose our final models based on performance and the clinical relevance of their predictors. We will then assess the true generalisation error of our final model44 in the test set. We will also summarise the final model's ability to predict survival at 1 year, 2 years and 5 years using the area under the receiver operating characteristic curve (AUC). Other regression models may also be considered in sensitivity analyses (ie, logistic regression models for the development of TB by time points).

We will also perform a descriptive analysis of the medical and demographic variables of interest in the complete data set. TB incidence will be calculated for all demographic and medical risk factors with corresponding 95% CIs.

Statistical analyses will be performed using R (V.3.2.5) and SAS/STAT (V.9.4, SAS Institute, Cary, USA). In general, missing and incomplete data will be excluded from analyses and the number of observations omitted from analyses due to missing data will be documented. When imputation is used, the method and extent will be reported. Observed p values of <0.05 will be described as statistically significant and 95% CIs will be provided for relevant parameter estimates. No corrections will be made for multiple inferences in this predictive modelling analysis.