Behavior Change and Digital Health Interventions for Improved TB Treatment Outcomes

Overview

Each year, 10.4 million patients are diagnosed with and 1.7 million people die from Tuberculosis (TB). Despite the availability of highly effective and accessible medications in the developing world where TB is endemic, the 6-18 month treatment regimen is often thwarted as patients fail to comply due to a lack of knowledge about the disease, desire for privacy, and/or stigma avoidance. Successful TB treatment is critical for reducing transmission, the selection of drug-resistant strains and treatment costs. Mobile health interventions promise to increase treatment success, especially in regions where directly observed treatment (DOT) is impractical. The most promising interventions attempted thus far employ a combination of SMS reminders and medication monitors. However, there is relatively little high-quality evidence on their impact, and what evidence there is shows mixed success. In Kenya, the burden of TB is among the highest in the world with a prevalence rate of 558 cases per 100,000 people. There is a great need for the development of alternative protocols, which reduce the costs of treatment and burden of adherence, and more effectively motivate patients to adhere to the program. A substantial and growing literature in the social sciences demonstrates the potential of behavioral interventions for generating large increases in contributions to public goods. Keheala, a feature-phone and Internet-based digital platform that uses Unstructured Supplementary Service Data (USSD) technology to register a patient's self-verification of medication adherence alongside support and motivation, based on proven techniques from the behavioral sciences, was shown in a 1,200-patient randomized controlled trial (RCT) to reduce the unsuccessful TB treatment outcomes in Kenya by two-thirds compared to the standard of care protocol. This 15,500 patient RCT will compare Keheala's scalability, cost-effectiveness and social impact to alternative interventions across diverse regions of Kenya.

Tuberculosis (TB) is the world's deadliest infectious disease. Each year, 10.4 million new individuals are diagnosed with TB, and 1.7 million are killed. Despite the availability of highly effective medications for treating TB in the developing world, lack of adherence to the treatment regimen remains the driving influence leading to multi-drug resistant (MDR) TB, morbidity and mortality. Conventional treatment is a lengthy process which depends heavily on patient adherence in seeking and then carrying out the prescribed treatment. After some time, the antibiotics typically eliminate symptoms of the disease, but the antibiotics themselves can make individuals feel quite ill. Additionally, individuals lack information, support, and suffer from a powerful community stigma. Stigma avoidance has been shown to drive work absenteeism, community outcasting and distant travel for health services. Because of these burdens, individual will become ill again - it also increases the risk that others will be infected, and that an antibiotic-resistant strain will evolve. There is therefore great need for the development of alternative protocols, which reduce the costs of treatment and burden of adherence, and more effectively motivate patients to adhere to the program. TB is spread through the air when people who are sick with TB excrete the causative bacillus, Mycobacterium tuberculosis mainly through coughing. Left untreated, a single patient can infect between 12 and 15 persons per year, a stark contrast to an Ebola patient who will only infect between 1.5 and 2.5 persons. When a patient adheres to the treatment regime, she makes it less likely that others will become sick, contributing to the health of her family and community. Adherence is thus a contribution to a public good--a personally costly action that benefits others. A substantial and growing literature in the social sciences demonstrates the potential of behavioral interventions for generating large increases in contributions to public goods, yet this potential has largely been left untapped in the treatment of TB. Keheala taps into this underutilized potential by developing a powerful platform for better engaging patients' sense of responsibility to their community in order to increase adherence. Keheala is a feature-phone and Internet-based digital platform that uses text message-like interactions to deliver behavioral interventions that have been demonstrated as remarkably effective in the social sciences literature on altruism, as well as in our proof of concept randomized controlled trial (RCT) administered in Kenya. Building on these early findings, the investigators now seek to: 1) Implement a large-scale, four-sided randomized controlled trial (RCT) that investigates large-scale, programmatic implementation across diverse and heavily burdened regions of Kenya, 2) compare cost-effectiveness with three practical alternative protocols, 3) assess the long-term social impact of the intervention and 4) link self-reported adherence with verifiable markers of medication consumption. Justification: Evidence of mobile interventions improving TB treatment outcomes falls short of expectations. There is need for more powerful and cost-effective interventions, for higher quality evidence of the impact of these interventions on treatment outcomes, and for studies specifically on TB. Keheala has been demonstrated in a large-scale RCT to substantially reduce the unsuccessful treatment outcomes of TB patients at 17 clinical sites around Nairobi County. Policy makers and funders now seek to 'stress-test' the Keheala intervention at scale across diverse and challenging environments, through its programmatic implementation and by comparing its cost-effectiveness to alternative protocols. The alternative protocols, an SMS-reminder intervention and a fully-automated, behaviorally informed mobile health intervention, are thought to be effective, less expensive and more easily scaled. Keheala is frequently compared to these interventions in conversation, but any evidence that suggests a benefit over the other is unsubstantiated today. Accordingly, the investigators propose a large-scale RCT, the gold standard for evidence-based research, to rigorously investigate questions of scalability and cost-effectiveness. General Objectives: The goal of the research is to compare health outcomes between the four treatment arms, with cost and cost-effectiveness being a secondary objective. Additional Qualitative Objectives: Of additional interest is whether adherence in the Keheala group is higher than in the control group. For this, the investigators will test urine samples in both groups. Within the Keheala group, whether those who self-verified have higher adherence rates will also be investigated. A separate analysis of health outcomes amongst MDR patients will be performed to understand the potential benefits for this sub-group of patients. Of note, the study was not powered to achieve statistical significance amongst this sub-group, noting their low prevalence amongst the TB population. Still, this is a sub-population that is of particular international interest. Accordingly, this group of patients will be investigated in the data for any potential insights. Measure and compare the health behaviors, health knowledge, attitudes, prosociality, employment and financial impact of the four interventions. The above topics will be investigated by surveying patients via mobile phones at the start and three months following completion of their treatment. Study population: Sampling procedure \& calculation of sample size: Health Outcomes: The investigators will employ a four-sided stratified RCT with 15,500 patients in total: a) 2,000 patients in the standard-of-care control group; b) 1,500 patients in the SMS intervention group; c) 6,000 patients in the SBCC intervention group (SBCC); 4) 6,000 patients in the Keheala intervention group. This methodology was selected for the following reasons: 1. RCTs are the gold-standard for providing evidence-based evaluation of medical interventions and recommended when feasible. 2. Keheala is a new program and cannot be sufficiently studied using existing data. 3. The researchers have an established record of designing largescale RCTs in real-world settings such as ours. This sample size was chosen to achieve 80% power in detecting improvements of 2.5% or more in the unsuccessful health outcomes. To do this calculation, Monte Carlo simulations of patient outcomes were performed based on results of the 2016 proof-of-concept RCT, knowledge of the effectiveness of SMS interventions and the potential of behavior-change communication for eliciting desirable behaviors. The investigators also powered the study to achieve 80% power with a 1% reduction in deaths from a base death rate in the control of 2.6%. This reduction in deaths is consistent with the reduction in our 2016 pilot. Our power calculation is probably a bit conservative, since base death rates are higher outside Nairobi. Urine Sampling: For urine sampling, 800 tests will be performed in a ratio of 3:1 (600 in the Keheala group, 200 in the Control group). This will enable detection of a 6% difference in average adherence at 80% power. Counties were grouped by region to ensure geographic representation in order to inform a national rollout. Within regions - Central, Coast, Eastern, North Eastern, Nairobi, Nyanza, Rift Valley and Western - a single county was selected for inclusion. Counties were selected for a combination of low treatment success rates, high caseloads, urban/rural representation, ethnic diversity and hard to reach areas. This process was performed in conjunction with Kenya's National Tuberculosis and Lung Disease Program (NTLDP) epidemiologist. The data used to select counties for participation is appended with brief explanations offered for each selection. The following counties, displayed by region, have been selected for inclusion: No. County Name Region 1. Kakamega Western 2. Kiambu Central 3. Kisumu Nyanza 4. Machakos Eastern 5. Mombasa Coast 6. Nairobi Nairobi 7. Turkana Rift Valley 8. Wajir North Eastern The investigators will stratify the data according to: county, gender, HIV-coinfection status, and language preference. Within each stratum, block randomization will be employed. In lay terms, this means that within a stratum, the intervention will be assigned according to a pre-specified, but random order. Note that since individuals are assigned to strata--and thus to study intervention--as they arrive, our procedure is an example of proportional stratification. The following random ordering was generated for block randomization: '4443331332314342133443243434414'. The numbers in the vector represent the interventions in the study: 1 represents the control, 2 the SMS reminder intervention, 3 the SBCC intervention, and 4 the Keheala intervention. The frequency of the numbers in this vector is proportional to the frequency in the study (4:3:12:12). The investigators will simply follow that ordering in all the strata. That is, the investigators will follow it for 'English-speaking HIV-coinfected males in county X', and follow it for 'English-speaking HIV-coinfected females in County Y', etc. Patients in all groups will receive a calling minutes top-up two-times: first, upon completion of a phone-administered (text) survey on the first day of enrollment, and again, three months after an outcome has been entered for the patient. Providing such compensation is common in Kenya. Intervention: All groups will continue to receive the standard-of-care protocol which includes diagnosis, the provision of medications - typically administered for a 1-2 week period at which point patients are expected to return to the clinical site - periodic on-the-ground follow-up by community health volunteers (CHVs)(at the discretion of local clinicians and CHVs) and nutritional support, as is sometimes provided. All groups will be periodically and consistently surveyed, once three days post-enrollment and, again, three months after completion of treatment. All phone-based interventions will be offered in either English or Swahili. This decision was made in conjunction with the NTLDP, which estimates 95% of patients can speak at least Swahili. For those who don't speak Swahili or English, clinicians will be trained to encourage patients to get help from an able family member. Clinicians will first consent patients, assisted by a 'Consenting and Enrollment Script' into the study and have them sign consent forms. Clinicians will provide patients a signed copy of the Informed Consent Form for their possession, while a second signed copy is kept and collected by SCTLCs. Keheala will collect signed consent forms from SCTLCs at Quarterly Cluster Meetings. Patients will be registered into our study by clinicians on their own cell phones. Clinicians will confirm over USSD that a patient consented before they are able to enroll the patient. A clinician enters in a patient's name, phone number, treatment time, language preference, gender and HIV status. After this point, patient's will be randomized by the system into one of the four study groups. At the end of the enrollment process, clinicians will be prompted to provide any needed materials to the patient according to the group they are randomized into. Patients in the SBCC and Keheala groups will be provided a 'Patient Flyer' that details how to access the platform. Patients in the Control and SMS Only groups will not receive any physical materials. Signed consent forms will be collected from clinical sites by sub-county coordinators and collected by study staff at quarterly regional meetings. Clinicians can request a resupply of materials across the Keheala platform. Urgent resupply requests will be served by courier while less urgent requests will be fulfilled at quarterly regional meetings with study staff disbursing materials to Sub-County TB/Leprosy Coordinators (SCTLCs) for distribution to clinical sites. Urine Testing Protocol: In order to validate self-reported adherence, the investigators will verify verification with GFC Diagnostic's IsoScreen urine test, which can be administered outside of a laboratory setting to identify Isoniazid in urine. The test has demonstrated high validity and reliability with a sensitivity and specificity of 95% and 98%, respectively. 800 IsoScreen tests will be ordered six to eight weeks in advance of desired sampling period allowing sufficient time for production, shipment and passage through customs. Our contacts at USAID in Kenya and at the NTLDP will help us to navigate the local procurement process. Only patients who have been verified as TB patients in TIBU (see 'Data Management' section) and who are still on treatment at the time of sampling (to be determined according to enrollment rates) will be included in the randomization process for urine sampling. Drug-resistant patients on conventional treatment (20 months) will be excluded from urine sampling as they do not receive Isoniazid in Kenya. Clinics with at least 100 patients enrolled will be eligible for the urine sampling procedure. Of these eligible clinics five to ten will be randomly selected to participate. Within these five to ten selected clinics enough patients for sampling will be randomly selected so that an estimated 800 tests can be performed (accounting for the likelihood of finding a patient). The investigators will first attempt to find these individuals at the clinical site over the course of a week. Those who do not appear at the clinical site will be pursued via home visits the following week. While there will be some sample selection, there are HIV studies in Uganda that show small differences between adherence as tested at clinic visits versus at unannounced home visits. Study staff will visit selected clinics for a week at a time and identify the randomly selected patients for urine sampling during the patient's weekly clinical visit. Those patients who do not appear at the clinical site will be visited at their homes, unannounced, the following week. Addresses will be retrieved from TIBU. If a patient is not found during the first-attempted home visit, they will not be revisited out of fear of changed-behaviors in anticipation of a follow-up visit (communicated by family members or friends). Study staff will be identifiable by their 'staff' shirts and will communicate their position as study team members. Patients will be reminded of their consent to participate in the study. Patient names, phone numbers, the date of the test (i.e. JUL 15 2010) and time of test (in military time) will be recorded. A patient's identity will be verified by photo ID or patient card. Patient and sample collector will initial the record to indicate accuracy and identities of those present. Patient phone numbers will be checked by USSD to ensure linkage within the database for future analysis. The following sampling procedure is then carried out: 1. A patient is provided a urine sampling container and requested to provide a sample. 2. Once returned, the study team member will fill a testing syringe with urine from the container and place onto plastic test tube. Holding the tube in both hands on a firm surface, both thumbs should be placed on the sides of the cap. The cap should be snapped on quickly. 3. The syringe is then compressed to expel its contents into the reaction tube. The sample can be mixed by flicking the tube vigorously for 30 seconds. Color change, if any, should happen rapidly. 4. Color determination should be made within five minutes of mixing the sample. Reading the sample after five minutes will give inaccurate readings. 5. Urine samples, container, testing vile will be disposed of locally after color indication is recorded. If there is no suitable means for local disposal, discarded materials will be collected in a sealed bag and disposed of at a later time in a suitable location. Data collection and reduction: County and sub-county TB coordinators visit each clinical site every one to three months and compile data from the paper TB registers into the NTLDP's electronic reporting system, TIBU. TIBU simultaneously generates a sub-county registration number by which each individual case can be identified and tracked. This number is recorded in the facility's TB Register. Clinicians are trained and incentivized to return to their phone's USSD portal to link study-registered patients with their sub-county registration number. The researchers will receive weekly TIBU exports that our computer system will automatically cross-reference to verify registered individuals are, in fact, TB patients according to TIBU. Only once this validation is completed are clinicians compensated for the patient he or she enrolled. Once a patient is registered in the system, they remain 'active' in the study until a corresponding clinical outcome is detected in the TIBU exports. At that point, the patient is automatically 'deactivated' from whichever intervention they are receiving and our primary data point is recorded. An effort (depending on what is logistically and financially feasible at the time) will be made to follow up with a sample of LTFU patients to understand the nature of their disappearance from treatment. As a backup, any individual started on TB medication is entered into a paper log at each health facility. Points of contact at each clinical site will be established by sub-county coordinators who will enroll local clinicians. Accordingly, data sharing will primarily happen electronically and remotely, as a backup, information can be shared by and in a worst-case scenario, Keheala will arrange to pick-up prepared documents with updates from the clinical sites. Platform usage data and analytics will be obtained from a dashboard created by our tech supplier. Qualitative data will be obtained from surveys digitally administered across the mobile platform and patients will be offered airtime for successfully completing surveys (x2). Free response questions will be categorized and summarized after collection. Urine sampling paper logs will be digitized when returning from the field and merged with our electronic database by patient phone number. Cost Data Collection: Cost analysis will focus primarily on costs of each intervention, based on materials (e.g. hardware, software), and personnel time-both training as part of rollout, and ongoing support to patients and programs. As the project budget covers all these, much of the necessary cost information will be available from detailed accounts from the project itself, and will be prorated/attributed to the number of patients covered. No changes in costs to patients are expected i.e. they will not be traveling more or less for visits, taking more or less time off work or family duties, etc. Similarly, it is not anticipated that other health system costs (e.g. health care personnel time, supplies, etc.) will be directly affected by any of the interventions. The investigators will also consider downstream cost effects with respect to any differences in the need for retreatment (i.e. as a consequence of unsuccessful outcomes). Statistical Analysis: In our main analysis, the investigators will compare outcome rates at the end of treatment for subjects in the intervention and control groups. This comparison can be done using a t-test or using an ordinary least squares regression (the latter was done in our power calculations), which can include controls for time trends, patient demographics, and clinic characteristics. Differences in treatment effects by individual characteristics, including MDR status, will also be considered. Finally, the relationship between verification rate (average days verified on the platform) and the primary and secondary outcomes will be analyzed. For the two groups with daily verification data (SBCC and Keheala) a more sophisticated (and slightly more statistically powerful) analysis of hazard rates will be employed. The hazard rate is estimated daily as the probability that a patient who is currently adhering continues to adhere. It is estimated using a logistical regression of whether the patient adhered on a given week on whether the patient adhered in all previous weeks and whether the patient is in an intervention group. The regression can also include controls for time period, patient demographics, and clinic characteristics. Note that the power calculations are based on the simpler regressions of final outcomes by intervention groups. Cost-analysis will examine the cost per patient of each intervention, while cost-effectiveness analysis will examine the differences in cost relative to differences in successful outcomes (i.e. incremental cost per additional cure/completion). If mortality differences are demonstrated, the investigators will also estimate costs per DALY averted. Urine Testing: If the sample develops a blue color, it is positive for isoniazid metabolites, and is therefore from a patient who is compliant with the treatment. A blue/purple color indicates that the drug was taken within the last 24 hours. A green color also indicates isoniazid metabolites, but the drug was probably taken about 48 hours ago. If the sample remains yellow, then the patient is not adhering to their daily treatment. Time stamps from sample collection will be compared with the patient's previous treatment time. If the time period since last treatment is consistent with the time-lapse indicated by the color of the sample, the individual will be indicated as adherent to treatment on that day. For individuals in the Keheala group, the accuracy of self-verification will be measured. i.e. do people who say they verified in the last 24 hours actually consume medication, or vice-versa. Using regressions to control for individual characteristics, comparisons of proportion of medication consumed will be made for Keheala vs. control groups. Within Keheala, people who say they verified in the last 24 hours and whether they actually consume medication more often will also be investigated. Study Duration: The entirety of the study will take place over a 30-month period. This time frame includes two and a half months of technological development, planning and approvals, one-month hiring and training local employees, testing the technology, training CTLCs and SCTLCs, followed by 24 months implementing the actual intervention. The RCT will be followed by a final one to three months of data collection and analysis. Report Preparation: At the completion of the project, the report will be prepared within a six-month time frame. The report will be prepared by Dr. Erez Yoeli from Yale University, with support from the listed researchers as needed. The report will present an analysis of the hypotheses listed as well as a discussion of the qualitative objectives. Study Closure Plan and Procedure: Patients will receive their intervention until a treatment outcome has been assigned. Following the assignment of a treatment outcome, patients will lose access to their intervention. A final digital survey will be administered to patients in accordance with the timing protocols stated above. Data will continue to be shared following the termination of operations until all patients have finished treatment and outcomes are determined. Patients who do not finish treatment within the planned 24-month period will be afforded access to their respective intervention as long as it is financially and logistically feasible. On July 31st, 2020, four months after the expected end of study operation, anyone who has not yet completed treatment will be marked 'INCOMPLETE' and their results considered separately. Analysis and report preparation will follow final results. \-------------------------- SAP Addendum - October 2019 The investigators would like to provide additional clarifications to the statistical analysis plan. Clarifying definitions of outcomes: The primary study outcomes are composed from treatment outcomes that are recorded by clinicians in the clinics' TB 'registers' according to World Health Organization guidelines. These outcomes include: cured (a bacteriologically confirmed individual whose sputum smear or culture was negative at month five or later), treatment completed (a TB patient who completed treatment without evidence of failure but with no record to show that sputum smear or culture results at month five or later, either because tests were not done or because results are unavailable), misdiagnosed (an individual who was originally diagnosed but subsequently reported as not having TB), transferred out (an individual who transferred to another clinic), died (an individual died during TB treatment), failed (an individual whose sputum smear or culture was positive at month five or later), loss to follow-up (an individual who did not start treatment or interrupted treatment for two or more consecutive months; abbreviated as LTFU). The investigators define the binary variable 'unsuccessful treatment outcome', which indicates whether an individual's outcome was any of: died, failed, or LTFU. The primary study outcomes are unsuccessful treatment outcomes, died, and LTFU. Clarifying the statistical tests: Individuals who were misdiagnosed or who transferred out will be omitted from the analysis. The number of individuals omitted will be reported. For the primary outcomes (unsuccessful treatment outcome, died, and LTFU), the investigators will perform a t-test comparing each intervention group to the control group. This analysis will be performed for all patients, restricting to bacteriologically confirmed patients, and restricting to MDR-TB patients. Note that the study is powered for comparisons that include all patients, and the investigators have selected the sample size with these comparisons in mind. I.e. the study is not powered for comparisons that restrict to MDR patients. The investigators will also fit logistic regressions to estimate the marginal effect of the interventions (I.e. a logistical regression of the outcome on three binary indicators, which equal one if the individual was included in the intervention in question, and zero otherwise). In some regression specifications, the investigators will include binary indicators for each clinic as a fixed effect and controls for individual characteristics collected from the TB registers and a short survey performed at the onset of treatment. Again, this analysis will be repeated restricting to bacteriologically-confirmed patients, and MDR-TB patients. For an example of how such an analysis was performed in the past, see: https://www.nejm.org/doi/full/10.1056/NEJMc1806550 Finally, to explore how the treatment effects different individuals, the investigators will fit "leave-out" and logistic regressions of unsuccessful treatment on individual demographics for the control, and use the results to predict likelihood of unsuccessful treatment for all individuals. The investigators refer to the resulting prediction as an individual's likelihood of unsuccessful treatment. Individuals will be sorted into quintiles according to this variable, and actual unsuccessful treatment by intervention group and quintile will be displayed. The same analysis will be performed for LTFU, and restricting to bacteriologically confirmed individuals. Addressing an Enrollment Randomization Error: Part-way through enrollment the investigators discovered and corrected an error in the randomization procedure. The investigators planned to randomize subjects into treatment groups according to a block randomization. Practically speaking, this randomization was implemented using a string made up of a randomly ordered sequence of the digits 1-4, where 1 represents control, 2 represents the SMS only treatment, 3 represents the SBCC treatment, and 4 represents the Keheala treatment. The appropriate number of each digit was included to ensure each treatment group would reach the pre-registered size once the total sample reached 15,500, then generated a random permutation of the string. It was: 4-4-4-3-3-3-1-3-3-2-3-1-4-3-4-2-1-3-3-4-4-3-2-4-3-4-3-4-4-1-4 In February, 2019, the investigators discovered that the last three digits of the string had accidentally been removed, and that the sample had, up until that point, been randomized according to the following string instead: 4-4-4-3-3-3-1-3-3-2-3-1-4-3-4-2-1-3-3-4-4-3-2-4-3-4-3-4 On February 20th, 2019 the string was updated to correct this error, and from that point, the original string was used. Subjects will continue to be enrolled with the original string until all groups reach the pre-registered size, even if some groups exceed this size as a result of the error. Changes to Urine Sampling Protocol: The procedure for conducting surprise urine samples yielded much lower than expected enrollment, due to the complexities of attending facilities and conducting surprise home visits. Consequently, the investigators updated the procedure twice to include additional facilities from which to select patients. The update was done consistently with the original procedure for selecting facilities. When the investigators originally chose facilities (October 2018), this was the procedure used: On October 30th, Patient Registration data was exported. The following counties had at least two facilities with \> 100 patients each OR at least one facility with \>200 patients: Kakamega, Machakos, Nairobi, Turkana These counties were organized by region and two regions were randomly selected (green) according to the lowest randomly generated numbers The first time facilities were added (April 2019), this was the procedure used: User and facility data was exported in early April 2019 Facilities with less than 10 active users were removed (n=13). Referral facilities (n=11) were removed as there is unlikely to be a high number of active patients receiving care there. Remaining facilities (n=38) were assigned a random number. Within a county, facilities were ordered for a visit from low number to high number. The originally randomly selected counties, Kakemega, Nairobi and Machakos will be visited first. If additional patients are needed for sampling, Kiambu and Mombasa facilities will be visited in accordance with the randomized order assigned. The second time facilities were added (August 2019), this was the procedure used: User and facility data was exported August 29th, 2019. Facilities with less than 10 active Keheala/Control users were removed (n=27). Referral facilities (n=11) were removed as there is unlikely to be a high number of active patients receiving care there. Remaining facilities (n=32) were assigned a random number. Within a county, facilities were ordered for a visit from low number to high number. The originally randomly selected counties, Kakemega, Nairobi and Machakos will be visited first. If additional patients are needed for sampling, Kiambu and Mombasa facilities will be visited in accordance with the randomized order assigned.