Therefore, this study would investigate the situation of pre-treatment loss to follow-up of persons with active tuberculosis within the Central African region using Cameroon as a case study. At the moment in Cameroon, the NTP does not recommend the searching and putting on treatment of all those that are lost after diagnosis, and does not also recommend TB treatment counselling of these patients before their test results are out from the laboratory. This is one of the main reasons why we went ahead with this study to determine the magnitude of this phenomenon in the first place, and the NTP will be advised to find solutions to this. This is the more reason why we equally characterized this group of patients for the Programme and healthcare providers so that they could target particularly those suspects who present with such characteristics. We hypothesized that the magnitude and risk factors for PLTFU may be different from that in other places. Therefore, in this study, the proportion of PLTFU, reasons for not initiating treatment and their risk factors were studied. This is the first study in Cameroon to look into the incidence of PLTFU of TB patients in Cameroon, thereby highlighting the problem of PLTFU versus programme performance. We aimed to conduct three studies with the following objectives:
2.1 Retrospective study objectives:
The study objectives included determining the incidence of PLTFU of patients with bacteriologically-confirmed pulmonary tuberculosis; and identifying risk factors associated with PLTFU of confirmed TB patients in the North West and South West regions of Cameroon, using routine TB program data.
2.2 Prospective study objectives:
Through prospectively counselling and tracing the newly diagnosed TB patients (i.e.
patients who were being diagnosed with TB for the first time), this study sought to
determine the incidence and risk factors of PLTFU of patients with bacteriologically-confirmed pulmonary tuberculosis; and importantly to identify reasons (from the
patient’s perspective) associated with PLTFU of bacteriologically-confirmed pulmonary tuberculosis TB patients in the North West and South West Regions of Cameroon.
2.3 Impact evaluation of patient counselling and phone reminder study objectives:
The counselling of patients and phone reminder in the prospective cohort study and the availability of the data from the retrospective study provide an opportunity to assess the impact of interventions implemented in the prospective study on PLTFU. We aimed to evaluate the impact of counselling of patients and phone reminder on the incidence of PLTFU using a before-and-after study design.
14
CHAPTER 3. METHODS
3.1 Overview
This study was conducted in two regions of Cameroon, the North West (NW) and South West (SW) Regions located in the southwestern part of the country (Figure 3). The study included 39 out of the total 40 TB diagnostic and treatment units (DTU) also known as TB basic management units. These sites are located in the majority English speaking area of Cameroon, with about 3.5 million inhabitants and at least 2000 TB cases detected annually from both regions. These DTUs are located in both rural and urban settings within the regions and are within public and private health facilities. This methods’ section will include three different study designs in view of the three studies: First, the retrospective study was used to understand the magnitude and determinants of PLTFU using routine program data under the business-as-usual setting. Second, the prospective study was designed to further understand the determinants of PLTFU by prospectively collecting information on sociodemographic factors (which were not available in routine data). It was also designed to identify the reasons of PLTFU through active tracing and calling the PLTFU patients in order to collect information on reasons for not initiating treatment. Lastly, the impact evaluation study is a post hoc analysis after the completion of the retrospective and prospective studies. We evaluated whether the patient counseling and phone reminder delivered in the prospective study would have an impact on PLTFU using a before-and-after study design.
3.2 Study population
The study included all consecutive bacteriologically-confirmed pulmonary tuberculosis patients diagnosed between July 1, 2015 and August 31, 2016 in 39 out of the total 40 DTUs of the NW and SW Regions during the study period with a total of 2160 bacteriologically-confirmed pulmonary tuberculosis cases. The mode of diagnosis of patients was mainly by
microscopy and GeneXpert. The month of January 2016 was however not included for data collection due to the investigators’ discretion. The retrospective study was conducted between July 1, 2015 and December 31, 2015 and 1174 cases were recorded. In the retrospective study, all diagnosed cases of all ages were included. While, the prospective study recorded 1060 cases and only those of age 21 and above were included. The prospective study was conducted between February 1, 2016 and August 31, 2016. The only exclusion criterion here was any patient that was transferred in from another DTU. The quasi-experimental study population included a total population of 2160 cases from the merging of the previous two study samples.
The patients were followed up for up to 30 days.
3.3 Data collection and management
A pre-prepared data collection form was used to collect information directly from the TB laboratory register to be cross-linked with the TB treatment register at each health facility by the primary investigator and trained data collection officers. All data collected were double-entered into a Microsoft Access database (Microsoft Corp, Redmond, WA, USA) to ensure data integrity. This information was securely saved in compliance with the rules and regulations of the Internal Review Board requirements and ethical protections. All the data collected were appropriately checked for accuracy and completeness, verified by data collectors and investigators for errors and missing information. All data collected for these three studies were processed and analyzed using Stata Statistical Software: Release 13.1 (College Station, TX: StataCorp LP.).
For the retrospective study, the data extracted from the registers included the patient’s socio-demographic information such as age, sex, residential address, sputum examination results and treatment status. Equally collected were the region of DTU’s location (NW versus SW), type
16
of health facility (public versus private), location of the DTU (rural or urban) and method of diagnosis.
For the prospective study, in addition to what were collected in the retrospective study, we used the pre-prepared data collection form to obtain information on socio-demographic and socioeconomic indices; and a semi-structured questionnaire to obtain their reasons for not initiating treatment. Before participating in the study, the patients were counselled about tuberculosis and the reasons for us conducting this study. Patients were told of the importance of completing the TB investigation and the need to start treatment immediately after diagnosis as is required by the NTCP. They were also informed that they would be called if not yet started treatment by the 8th day to remind them to come and initiate TB therapy. Finally, an informed consent was given by the index patient or a care-giver to participate in the study. The socio-demographic and socioeconomic indices obtain were educational level, marital status, employment status, financial status and other health-related conditions. The reasons (i.e. the patient’s perspective only) for not initiating treatment earlier than expected was collected only from those who had not started treatment using an interviewer-administered semi structured questionnaire through phone calls or messages through other patients or community health workers. At most, three call attempts were made to reach patients after 7 days post-diagnosis (on days 8th, 9th and 11th) if he or she had not initiated treatment. Once the patient was reached, reasons for not initiating treatment were collected and the study on that particular case ended at that point with no further calls or search; however, all patients were followed-up passively for 30 days (a month). Additionally, these forms of communication also served as a means of reminding patients to get back to care.
The quasi-experimental study population included a total of 2160 bacteriologically-confirmed pulmonary tuberculosis cases from the retrospective and prospective study groups by merging the two databases.
3.4 Measurement of pre-treatment loss to follow-up
A PLTFU case was defined as any bacteriologically-confirmed pulmonary TB patient that was recorded in the laboratory register, but was not placed on TB treatment at that particular facility within 7 days. The 7-day cut off, though arbitrary, was decided after reviewing the various studies already conducted on PLTFU across the world especially in Africa with consideration of the field knowledge of the most common treatment initiation day in these two regions of Cameroon.
In addition to PLTFU, we also measured the treatment delay by calculating the time taken for a patient from the date of first positive acid-fast bacilli or GeneXpert result to the date of starting tuberculosis therapy (i.e. time to treatment). The time to treatment information would not be affected by the current cut off definition of PLTFU.
3.5 Measurement of the determinants of pre-treatment loss to follow-up
Information on potential determinants of PLTFU was obtained from the structured questionnaire designed for this study, see Section 3.3 for details. One key determinant of PLTFU in this study was the geographical access to DTUs, which was evaluated using travel distance and travel time between patient’s residence and DTU’s location by means of both global positioning and geographic information systems tools. We
obtained the geocodes of the DTUs using a global positioning system tool (GPS coordinates). The geocodes of the patients’ residences were obtained by matching the patients’ addresses through online maps (MapQuest, OpenStreetMap, Google Maps, HERE WeGo) and other relevant websites such as http://latitude.to/,
http://nona.net/features/, www.worldplaces.net and www.geographic.org. Travel distance and travel time calculations were estimated using an application programming
18
interface (API) code that was freely donated by HERE map developer to be used in Stata data analysis software.
3.6 Statistical analysis
3.6.1 Retrospective study
Descriptive analyses of study participants were done by tabulating the frequency distribution.
The proportion of PLTFU among all participants was calculated for PLTFU.
Maps showing the locations of patients (i.e. those who initiated treatment and those who had not) and that of the DTUs were produced using Quantum Geographic Information System (QGIS 2.18) software, Las Palmas, Spain.36
Univariable and multivariable logistic regression analyses were conducted to analyze whether geographical access and other possible determinants were associated with PLTFU, with the estimation of the crude and adjusted odds ratios (ORs) and their 95% confidence intervals (95% CI).
Our analyses of PLTFU did not consider the variation in treatment delay among those who were eventually treated. We therefore conducted a separate time-to-event analysis to investigate the determinants of treatment delay using univariable and multivariable Cox proportional hazards model. In the Cox regression model, survival data was constructed and
‘initiated treatment’ was set as ‘failure = 1’ while ‘not initiated treatment’ as ‘censored =0’.
In addition to geographical access, we considered other determinants of PLTFU in the multivariable analyses, including age, sex, location of DTU, region of DTU, health facility type and method of diagnosis.
We also used restricted cubic spline regressions with 3 knots to model the dose-response relationship between travel distance and the probability of PLTFU using its odds ratios on
the Stata statistical software. Three knots were used because PLTFU did not change quickly over travel distance, though we also applied four and five knots in the sensitivity analyses.37
3.6.2 Prospective study
For the prospective study, descriptive analyses of study participants’ characteristics in this prospective study were done by tabulating the frequency distribution; and the proportions of PLTFU for the characteristics of the participants were calculated.
Geographical access and other possible determinants of PLTFU were also analyzed by conducting both univariable and multivariable logistic regression analyses as was done in the retrospective study. In addition to geographical access, age, sex, location of DTU, region of DTU, health facility type (public versus private) and method of diagnosis (smear microscopy versus GeneXpert); we further considered other information collected in the prospective study, including marital status, level of educational, employment status, monthly salary, transportation cost from patient’s home to health facility, time taken to travel from home to DTU, any history of doing a TB test, awareness of TB result after TB test, awareness of HIV status, duration of last HIV test done and presence of co-morbidity. A separate time-to-event analysis was conducted to investigate the determinants of treatment delay using univariable and multivariable Cox proportional hazards model.
3.6.3 Impact evaluation of patient counselling and phone reminder study
A post hoc quasi-experimental (before-and-after) analysis was conducted to evaluate the impact of patient counselling and phone reminder (an intervention delivered in the
prospective study) on the reduction of PLTFU. The period of the retrospective study, (July 2015–December 2015) was used as the baseline reference group, and the period of the prospective study, (February 2016-August 2016) was then used as the intervention group.
We conducted univariable and multivariable logistic regression analyses to estimate the
20
crude and adjusted odds ratios (ORs) and their 95% confidence intervals (95% CI) between the intervention and PLTFU. We adjusted for other potential determinants of PLTFU in the multivariable analysis, including age, sex, location of DTU, region of DTU, health facility type, geographical access, and method of diagnosis. A similar Cox regression analysis was conducted to estimate the association between the intervention and time to treatment initiation.
In addition, a regression discontinuity analysis was conducted to account for any linear underlying time trend of PLTFU during the study period. This was done by first visually plotting the proportion of PLTFU during the whole study period, including the baseline period and the intervention period. We then fitted two logistic regression models to estimate the effect of intervention.
The first model included only the intervention effect (Int) and assumed no underlying time trend:
Model 1: Logit (p) = α + 𝛽1∗ 𝐼𝑛𝑡
The second model accounted for a linear time trend of PLTFU and allowed the trend to be different before and after the intervention. This was done by including the intervention effect (Int), a linear time trend (Time), and an interaction term between the intervention and time.
Model 2: Logit (p) = α + 𝛽1∗ 𝐼𝑛𝑡 + 𝛽2∗ 𝑇𝑖𝑚𝑒 + 𝛽3∗ 𝐼𝑛𝑡 ∗ 𝑇𝑖𝑚𝑒
3.7 Sample size and power estimation
In estimating the sample size of this cohort study with a dichotomous outcome from a single sample, we took into consideration the following assumptions:
1. From each of these two regions, about 1100 cases of bacteriologically-confirmed pulmonary tuberculosis cases will be diagnosed each year based on historical data;
consequently 550 diagnosed cases would be expected within 6 months of data collection with at least 90 cases per month.
2. An estimated proportion of PLTFU among positive cases was 12% from each region, and was estimated from previous studies and field experience. Therefore, 460 cases from each region (a total of 960) at least 1000 diagnosed cases of tuberculosis from both regions will be required to give this study a power of 95% confidence interval of between 9.4 and 15.0%, two-sided significance level (α) = 0.05.
3.8 Ethical clearance
Ethical clearance was obtained from the Cameroon Baptist Convention Health Services Institutional Review Board (CBCHSIRB), the Cameroon National Ethical Committee of Research for Human Health (CNECRHH) and the National Taiwan University Hospital Research Ethics Committee (NTUHREC). Administrative approvals were sought from the Cameroon Ministry of Public Health through the National Tuberculosis Control Programme (NTCP), the Regional Delegation of Public Health and hospital authorities. The interviewee signed an informed consent before participating in the study.
22