Predictors of mortality and survival probability distribution among patients on tuberculosis treatment in Vihiga County, Kenya

Background Tuberculosis (TB) related mortality remains a serious impediment in ending TB epidemic. Objective To estimate survival probability and identify predictors, causes and conditions contributing to mortality among TB patients in Vihiga County. Methods A cohort of 291 patients from 20 purposively selected health facilities were prospectively considered. Data was obtained by validated questionnaires through face-to-face interviews. Survival probabilities were estimated using Kaplan-Meier method while Cox proportional hazard model identified predictors of TB mortality through calculation of hazard ratios at 95% confidence intervals. Mortality audit data was qualitatively categorized to elicit causes and conditions contributing to mortality. Results 209 (72%) were male, median age was 40 (IQR=32-53) years while TB/HIV coinfection rate was 35%. Overall, 45 (15%) patients died, majority (78% (log rank<0.001)) during intensive phase. The overall mortality rate was 32.2 (95% CI 23.5 - 43.1) deaths per 1000 person months and six months' survival probability was 0.838 (95% CI, 0.796-0.883). Mortality was higher (27%) among HIV positive than HIV negative (9%) TB patients. Independent predictors of mortality included; comorbidities (HR = 2.72, 95% CI,1.36–5.44, p< 0.005), severe illness (HR=5.06, 95% CI,1.59–16.1, p=0.006), HIV infection (HR=2.56, 95% CI,1.28–5.12, p=0.008) and smoking (HR=2.79, 95% CI,1.01–7.75, p=0.049). Independent predictors of mortality among HIV negative patients included; comorbidities (HR = 4.25, 95% CI; 1.15-15.7, p = 0.03) and being clinically diagnosed (HR = 4.8, 95% CI; 1.43-16, P = 0.01) while among HIV positive; they included smoking (HR = 4.05, 95% CI;1.03-16.0, P = 0.04), severe illness (HR = 5.84, 95% CI; 1.08-31.6, P = 0.04), severe malnutrition (HR = 4.56, 95% CI; 1.33-15.6, P = 0.01) and comorbidities (HR = 3.04, 95% CI; 1.03-8.97, p = 0.04). More than a half (52%) of mortality among HIV positive were ascribed to advanced HIV diseases while majority of (72%) of HIV negative patients died to TB related lung disease. Conditions contributing to mortality were largely patient and health system related. Conclusion Risk of TB mortality is high and is attributable to comorbidities, severe illness, HIV and smoking. Causes and conditions contributing to TB mortality are multifaceted but modifiable. Improving TB/HIV care could reduce mortality in this setting.


Introduction
Whereas tuberculosis (TB) is preventable and curable, it remains a leading cause of mortality from single in-fectious disease agent 1 . In 2019, an estimated 10 million people fell ill with TB globally and a total of 1.7 million died 2 . More than two thirds of TB cases occur in southeast Asia (44%) and Africa (25%) while lower incidences are observed in the Eastern Mediterranean region (7%), the European region (3%) and the region of the Americas (3%) 2 . Although Kenya has been removed from the list of high burden drug resistant TB countries, it remains one of 30 high TB and TB/HIV burden countries globally, with Gabon and Uganda joining the list 3 . Kenya, has an overall TB prevalence of 426 per 100,000 population 4 and the disease is responsible for 6.3% annual deaths and total Disability-Adjusted Life years (DALYs) of 4.8% 1,5 . Although Kenya reported a slight increase in treatment success rate (TSR) for all forms of TB from 82.4% for the 2017 cohort to 84% for 2018 cohort 6 , these rates are still below global TSR target of 90%. Additionally, there was an increase in TB death rate from 6.3% to 6.5% in 2017 and 2018 cohorts respectively. During the same period, 13% of TB patients died in Vihiga County, making it second among counties with highest TB related mor-tality in Kenya 7 . Understanding factors associated with increased risks of mortality is important in prioritization of key interventions and target groups.
The WHO's 'End TB Strategy' and the United Nation's Sustainable Development Goals (SDGs) share a common goal of ending global TB epidemic 8,9 . The milestones and targets include a 90% reduction in TB deaths and 80% reduction in TB incidence by 2030 and 95% reduc-tion in TB deaths and 90% reduction in TB incidence by 2035 compared with 2015 figures 9,10 . However, the slug-gish progress towards achieving these targets across high TB burden countries such as Kenya remains the biggest hurdle 11 . This is despite rigorous intervention programs such as use of molecular and culture methods for TB diagnosis, use of short-course fixed dose combination drugs, nutritional support, TB/HIV collaboration and routine follow up of patients 12 . Although several studies [13][14][15][16][17] concur on the need for a person-centered interven-tion integrated within a multi-sectoral strategy, there still exists scanty information on reasons for often observed survival distributions and causes of increased risk for mortality among patients on TB treatment. Such infor-mation is useful for more accurate prediction of risks and occurrence of mortality.
Previous studies have associated TB mortality with female gender 13 , male gender 14 , bacteriological unconfirmed TB, advanced age 15 , comorbidities 16,18,19 , behavioral characteristics 16,20 and HIV 17,21,22 . However, routine surveillance data may not provide sufficient variables to assess survival probably patterns and analyse the influence of broader demographic, socio-economic and clinical factors on mortality of TB patients 23,24 hence may not be applicable to the local setting. Through prospective follow up design, this study evaluated survival probability distributions and identified predictors, causes and conditions contributing to mortality among TB patients in Vihiga County. Sub group analysis for HIV positive and HIV negative patients was also conducted. Findings of this study have implication of better understanding TB epidemiology and accurate prediction of risks and occurrence of mortality. This can permit cost effective intervention strategies and heightened surveillance in high TB burden settings.

Methods and procedures Setting
The current study was conducted in Vihiga county which is located in Western region of Kenya and has a popula-tion of 590,013 25 . The county has consistently reported high rates of mortality among patients on TB treatment 6,7,26 , hence it provided a robust context for the present study. The county has four TB treatment zones; Emuhaya, Vihiga, Sabatia and Hamisi. This study was conducted among TB patients diagnosed and followed up in twenty selected health facilities which account for 85% of TB cases annu ally in the county 6,7,26 .

Diagnosis, treatment and follow up of drug susceptible TB
In Kenya, TB diagnosis, treatment and follow up is guided by the integrated guideline 12. Bacteriological confirmation of TB is achieved through molecular, phenotypic and radiological techniques. All cases of drug susceptible TB except TB affecting meninges or bones and drug re-sistant TB are treated with chemotherapy comprising of two months (intensive phase) with rifampicin (R), isoni-azid (H), pyrazinamide (Z) and ethambutol (E) (RHZE) followed by four months (continuation phase) with ri-fampicin (R) and isoniazid (H) (RH). After preliminary assessment of patients, drugs are administered orally through fixed dose combination and the dosage is determined by patients' body weight. Most patients are treated on ambulatory basis. Patients are required to visit health facilities weekly during intensive phase and twice monthly during continuation phase, during which they are assessed, counseled and given the medicine. Treatment outcome is assigned immediately after treatment completion or after occurrence of an event that constitutes termination of treatment such as death or treatment interruption.

Study participants
This study included TB patients 15 years and older, duly diagnosed with TB, notified in treatment registers and the national electronic data base (TIBU) and started on TB treatment. Patients with TB of the bones and joints, TB meningitis and drug resistant TB were excluded from the study.

Study design
Observational cohort study was conducted among TB patients in Vihiga County, enrolled between June and December 2019. After obtaining baseline data, each patient was followed up through routine visits until s/he completed treatment, died or was lost to follow up.

Sample size
Sample size for the current study was calculated using Cochrane equation 27. The accessible population for this study was 850, the average number of notified TB cases annually in Vihiga County between 2012 to 2018 7,28 . Since sample size for this study (384) exceeded 5% of the population size (850), a correction formula was calculated [n = 384/ 1+ [(384 -1)/850] to yield 265. Ten percent (10%) of 265, (26) was added to compensate for anticipated drop-outs, hence the sample size (n) was 291.

Sampling procedure
The study area and twenty health facilities that account for 85% of TB cases annually were purposively selected, five from each of the four TB treatment zones. The sample was allocated to the four TB treatment zones and then twenty health facilities proportional to their annual contribution of TB patients. Within facilities, simple random sampling method was used to select eligible patients into the study.

Data collection
Structured questionnaires were administered to eligible TB patients through face-to-face interviews by twen-ty trained research assistants, each attached to the par-ticipating health facility. The questionnaires comprised closed-ended questions covering patient demographic, socio-economic and clinical characteristics. Follow up data and TB treatment outcome information were also in-cluded. Additional data focusing on HIV and comorbid-ities were captured from patients' clinical records while mortality audit data from patients who died was obtained using the national TB mortality audit tool.

Variable definition
The outcome variable for this study was all cause mortality (death), 'yes' or 'no'. The total follow-up time for each patient was 180 days, number of days from treatment initiation until completion of treatment. Patients who completed treatment and those who interrupted treatment were censored. Time until event occurred was defined as time in days from treatment initiation to death. Predictor variables included patients' demographic, socio-economic and clinical characteristics. Alcohol and smoking statuses were self-reported while alcohol Use Disorder Identification Test (AUDIT) scoring system was used to assess alcohol use 29 . Severely ill patients were defined as clinically very sick with at least respiratory rate > 30/min, temperature > 39°C, heart rate > 120/min and unable to walk unaided. Comorbidities in the current study includ-ed all other underlying conditions apart from HIV infection.

Statistical analysis
Quantitative data was uploaded to R for statistical analysis. Standard descriptive statistics such as frequencies, proportions, mean, median, range and standard deviation were calculated to demonstrate the demographic, socio-economic and clinical characteristics of TB patients and characterize their probability distributions. Survival analysis was used due to their usefulness in handling time to event outcomes and censored observations 30 . Kaplan Meier (KM) estimator was calculated to obtain univariable descriptive statistics for survival data, including estimation of survival probabilities by patient characteristics. The Log rank (Mantel-Cox) test for the equality of survival distributions was used to analyse the significance of survival differences of TB patients by their characteristics. Variables with p value < 0.05 and universal variables such as age and sex were included in multivariable analysis; Cox proportional hazard (CPH) model, was fitted to identify predictors of all-cause mortality through calculation of hazard ratios at 95% confidence interval (CI). Due to high TB/HIV coinfection, difference in mortali-ty and difference in clinical characteristics between HIV positive and HIV negative TB patients, the sample size was broken down into two subgroups based on their HIV status. For subgroup analysis, Kaplan Meier (KM) estimator was calculated to estimate survival probabilities by patient characteristics among HIV positive and HIV negative subgroup. CPH model, was fitted within each sub group to identify predictors of all-cause mortality through calculation of hazard ratios at 95% confidence interval (CI). Before fitting the covariates into the CPH model, proportional hazard assumption was checked by plotting Schoenfeld residuals against time to test for independence between time and residuals. Any covariate that violated the assumption was stratified. For variable analysis p-value <0.05 was considered significant. Data from mortality audit was qualitatively categorized to elicit causes and conditions contributing to mortality.

Ethics statement
The current study was approved by the Maseno Univer-sity Ethics Review Committee (MUERC) (Ref: MSU/ DRPI/MUERC/00707/19) and the National Commission for Science, Technology & Innovation (NACOSTI) (Ref: 192517) and was conducted according to Helsinki's declaration. Written informed consent was obtained from all participants and confidentiality was ensured throughout the study.

Predictors of all-cause mortality among TB patients
Before fitting the covariates into the multivariable cox model, proportional hazard assumption was checked by plotting Schoenfeld residuals against time to test for independence between time and residual. BMI category was found to significantly (2.1%) differ from zero at the 5% significance level, therefore, the final model was corrected by stratification of "BMI category" covariate. After simultaneously controlling for the potential predictor variables (Table 3), TB patients with underlying comor-bidities were almost 3 times more likely to die (HR = 2.72, 95% CI; 1.36-5.44, p = 0.005) compared to patients with-out comorbidities. Besides, patients who were severely ill during treatment initiation were five times more likely to die (HR = 5.06, 95% CI; 1.59-16.1, p = 0.006) compared to clinically stable patients. TB and HIV coinfected patients were two and a half times more likely to die (HR = 2.56, 95% CI; 1.28-5.12, p = 0.008) compared to the HIV negative patients, while those who smoked were almost 3 times more likely to die (HR = 2.79, 95% CI; 1.01-7.75, p = 0.049) compared to non-smokers. .0, P = 0.04) compared to non-smokers. Also, severely ill patients were almost 6 times more likely to die (HR = 5.84, 95% CI; 1.08-31.6, P = 0.04) compared to clinically stable while severely malnourished patients were five-times more likely to die (HR = 4.56, 95% CI; 1.33-15.6, P = 0.01) compared to normally nourished. Besides, patients with comorbidities were three times more likely to die (HR = 3.04, 95% CI; 1.03-8.97, p = 0.04) compared to patients without comorbidities.

Causes of death among HIV Negative and HIV Positive TB treatment patients
Mortality audit was conducted for 45 patients who died while on TB treatment. Specific causes of death were categorized based on patients' HIV status and are sum-marized in Table 4. Majority (72% (TB pneumonia = 44%, lung collapse = 17%, lung fibrosis = 11%)) of HIV negative TB patients died due to TB related lung disease while more than a half (52%) of deaths among HIV positive patients were attributable to advanced HIV disease. Sixteen percent of deaths among HIV positive patients were due to opportunistic infections while 22% of mortality in this group were linked to TB related pneumonia.  Figure 3 shows conditions contributing to death, but not related to the disease or condition causing it, categorized by HIV status of patients. Forty-four percent of mor-tality among HIV negative patients were attributable to delayed diagnosis while a third died due multiple patient and health system factors. For the HIV positive mortality was exacerbated by poor adherence to ART (37%) and delayed diagnosis (33%).

Discussion
Ending TB epidemic requires substantial reduction in TB incidence and mortality 8 . This study indicates a high mortality rate among patients on TB treatment predominantly occurring during the intensive phase of treatment.
The cumulative incidence of mortality of 15% in the study area is higher than the national average of 6.4% 6 , Kilifi County (5.5%) 31 and Tanzania (3.6%) 15 . Similar or higher rates have been observed in other high TB burden countries such as Uganda (15%) 32 , Nigeria (16.6%) 33 and Zimbabwe (20%) 22 . Although only slightly more than a third of TB patients had one or more comorbid condition(s), patients with comorbidities were associated with significantly reduced survival probability and were independently associated with all-cause mortality including among HIV positive and HIV negative subgroups. Previous studies 16,18,19 have similarly demonstrated that comorbidities including diabetes mellitus, malignancies, chronic respiratory conditions, mental health illnesses, liver diseases and chronic kidney diseases are common among TB patients and significantly increase the risk of mortality.  43,44 have similarly attributed death among TB patients to non-TB causes. Comprehensive HIV care, treatment and prevention presents a unique opportunity for improved TB treatment success. There is also need to further understand reasons for reduced survival probability among TB/HIV coinfected patients previously on TB preventive therapy (TPT). Among the HIV negative group, patients with comorbidities and those clinically diagnosed were associated with increased risk of mortality. From mortality audit, TB related lung disease was the main cause of death. Conditions contributing to death in HIV negative subgroup are consistent with patients' socio-behavioral characteristics and the health system gaps which are modifiable through timely and accurate diagnosis, management of comorbidities, patient education and adherence counselling.