Cost-effectiveness of a Population-based Active TB Control Program for Tuberculosis, Lima, Peru
Cost-effectiveness of a Population-based Active TB Control Program for Tuberculosis, Lima, Peru
The purpose of this study is to determine the effectiveness and cost-effectiveness of a new population-based active case-finding program among adult household contacts of new infectious TB cases to detect active TB cases in the largest district, Lima, Peru.
The Ministry of Health TB program of the District SJL (DISA NTP) is undertaking new public health intervention that involves the active evaluation of household contacts. We are collaborating to roll this intervention out as a stepped wedge pragmatic randomized controlled trial in order to evaluate the effectiveness and cost-effectiveness of this intervention.
Intervention Model
The intervention is rolled out across all clusters (or health centres) in the form of a stepped wedge design, that is a unidirectional crossover, such that all clusters start in the control arm and are then randomized to the time point of cross over to the intervention arm. This continues in steps until all clusters (or health centres) have crossed over to the intervention arm. The randomization of health centres is to the timepoint at which they will cross over to the intervention arm. This randomization was stratified by clinicTB rates <100/100,000, TB rates 100 to 200/100,000, and clinics with TB rates >200/100,000, to ensure balance of clinic sizes randomized to the intervention arm at each time point.
Power analysis
Given that the number of clusters, or health centres (n=35) is fixed, a power analysis was undertaken to estimate the smallest detectable difference between the overall yields of the active intervention (ACF) and the passive case finding (PCF) in the current study. The power analysis for stepped wedge clustered designs must be modified to account for the stepped implementation and the variation in cluster sizes we are likely to observe. In a parallel study designed to achieve a specified level of power, the same population in a stepped wedge design will have less the specified level of power, and the standard normal deviates should be increased by an inflation factor. Using a similar approach and accounting for stepped wedge design and the varying cluster sizes within the sample, in comparison to the expected baseline 1% case detection among household contacts we anticipate in passive case finding, we should have a power of greater than 90% detect differences between 3% and 4% ( with overall yields of ACF with 4 or 5%) between the intervention and control arms. Based on previous studies, we anticipate this should be adequate to determine an effect that would be of clinical and programmatic importance.
The study will be undertaken in district clinics over the course of 18 months (3 baseline and 15 months of the ACF program phase-in). In 2010, DISA NTP reported 1,871 TB cases within 35 clinics over 12 months (unpublished Ministry of Health data). We anticipate median of 5 household contacts per TB case (range 3-7) . In previous studies within the district, response rates to index case enrollment to studies in the district have been over 80%. In a study recruiting 60 Multidrug resistant (MDR) TB index cases and 80 randomly selected drug-sensitive TB controls in SJL, 92% of case and 98% of controls selected agreed to participate in the study. Therefore, given our current proposed roll out over 15 months, if we conservatively estimate 70% participation and 3 household contacts per home we would anticipate approximately 1900 index TB cases and approximately 5700 household contacts.
Data extraction and management
DISA NTP routinely collect data on household contacts evaluated in public health forms within TB case histories. Similarly, during home visits, additional ACF forms on household contacts will be collected. The new ACF TB forms are public health forms that DISA NTP has developed, however with input from the study investigators, both to ensure data completeness and quality and to incorporate any data on time and cost of the program. The study investigators will provide a role in pilot testing and validation of the data collection tools and program procedures. DISA NTP staff will conduct all home visits, all evaluations TB cases and contacts and related data collection. Currently all NTP data at the clinic level are paper-based and maintained in individual TB case charts.
Study specific teams (university-based) will audit and extract the routine public health and the active intervention TB forms. This strategy will ensure a real-life pragmatic perspective of the program using DISA NTP staff to implement the program in the context of their regular TB program, however provide additional monitoring and digitizing of data for evaluation. Dedicated study field workers will extract PCF and ACF data prospectively from the original TB case histories, household contact evaluation and ACF intervention (where applicable) forms within index TB patient charts at DISA NTP clinics. Information from the case histories for each TB patient's household contacts will be collected in the first month, at 3 months and 6 month time periods. The extraction method will include where possible and allowed photocopy of the relevant public health forms. These forms are then summarized in study data extraction forms and entered into a single secure database. Data are verified by a supervisor comparing extraction forms to the chart forms, and the database electronic records to the extraction forms.
Analysis
The primary outcome of interest is the overall yield of secondary TB cases from either the new active program compared with the passive routine program and associated costs of each. Univariate analyses will be performed to describe population characteristics. Analysis will be undertaken according the principles of intention-to-treat (ITT), using data collected from individual TB case records at randomized clinics, such that contacts detected as secondary cases will be assigned to the intervention or comparator in their clinic at that time period. A marginal logistic regression model will be estimated using Generalized Estimated Equations (GEE ). The GEE will be modeled using a logit link function, and a robust variance estimation which is robust to misspecification of the correlation structure. Analysis will be at the unit of the individual accounting for the clustering by clinic. Covariates included in the model will include individual level age, gender, previous TB treatment, index TB case smear positivity gradient and clinic level covariates. The treatment effect will be an odds ratio of the time to case detection in the intervention group (ACF) in comparison to the comparator (PCF).
Inclusion Criteria:
Exclusion Criteria:
lena.shah@mail.mcgill.ca514-357-4913