Longer hospital stay is associated with higher rates of tuberculosis-related morbidity and mortality within 12 months after discharge in a referral hospital in Sub-Saharan Africa

All patients admitted to the medical wards between May 01 and July 31, 2010 at a large tertiary referral hospital in Botswana, were enrolled in this observational cohort. With a base population of approximately 350,000 persons, this hospital serves as the main referral centre for the Southern region of the country [32]. Wards are “open” (Figure 1) and divided according to the sex of the patients.

Data were collected on standardized forms on a daily basis. Every day, nurses reassess the patients’ diagnosis and clinical status as well of cubicle occupancy and staffing and patients relocated accordingly. As most of such patient relocation occurs during the morning nursing shift, all data were consistently collected at the same time of the morning and evening, after the patient re-location was finalized. The number of patients per cubicle, their position, environmental factors was collected at this time. More specifically, the following variables were collected from the medical records:

Every 3 months for a total of 12 months, data from each enrolled patient were extracted from paper medical records and electronic databases at the hospital, TB clinics, Botswana National Tuberculosis Program, the Botswana National Tuberculosis Reference Laboratory, HIV clinics, and the Botswana National HIV Program. Additionally, the National Death Registry was searched for details about patient mortality, date of death, and hospital readmissions. Data abstraction was focused on the identification of new episodes of TB, episodes of TB recurrence or relapse, overall mortality and TB-associated mortality. Patients unable to be contacted by any of the above methods were considered ‘lost-to-follow-up’. For this analysis, ‘untraceable’ patients were considered to be alive since their names could not be found in the National Death Registry. Due to the observational nature of the design, patients were never contacted during the entire duration of the study.

Given the different rates of progression to active PTB after the exposure of HIV-infected and HIV-uninfected patients, all analyses were performed separately by HIV infection status. Tuberculin conversion was not used as an outcome given the extremely high rates of positivity in our population and the almost universal exposure to TB among adults in our setting. Thus, our main outcomes were TB incidence, all-cause mortality and TB-associated mortality during the follow-up period. We used the World Health Organization (WHO) recommendation to define our TB cases [33, 34]: a) Case of TB: A patient in whom TB was confirmed by bacteriology or diagnosed by a clinician. b) Definite case: A patient with positive culture for Mtb or a patient with two sputum smears positive for acid-fast bacilli (AFB+). c) Pulmonary case: A patient with tuberculosis disease involving the lung parenchyma. d) Smear-positive pulmonary case: A patient with one or more initial sputum smear examinations (direct smear microscopy) AFB+; or one sputum examination AFB + and radiographic abnormalities consistent with active pulmonary tuberculosis as determined by a clinician. e) Smear-negative pulmonary case: A patient with pulmonary tuberculosis not meeting the above criteria for smear-positive disease. Diagnostic criteria included: at least two sputum smear examinations negative for AFB; and radiographic abnormalities consistent with active pulmonary tuberculosis; and a decision by a clinician to treat with a full course of ATT; or positive culture but negative AFB sputum examinations. f) Extrapulmonary case: A patient with tuberculosis of organs other than the lungs (e.g. pleura, lymph nodes, abdomen, genitourinary tract, skin, joints and bones, meninges). Diagnosis was based on one culture-positive specimen, or histological or strong clinical evidence consistent with active extrapulmonary disease, followed by a decision by a clinician to treat with a full course of ATT. A patient in whom both pulmonary and extrapulmonary TB was classified as a pulmonary case. g) New case: A patient who never had treatment for tuberculosis. h) Re-treatment case: A patient previously treated for TB, who started on a re-treatment regimen after previous treatment had failed, who returned to treatment having previously defaulted, or who was previously declared cured or treatment completed and was diagnosed with bacteriologically positive TB (relapse).

Descriptive statistics were determined by standard methods. Groups were compared using Fisher’s exact or Wilcoxon-rank-sum tests as appropriate. Odds ratios were calculated for all factors potentially associated with a final diagnosis of PTB. We controlled for possible confounding or effect-modifying factors by using logistic regression. We determined potential confounders a priori based on estimation of their significance as epidemiologic factors during the preliminary crude analysis (significant at p ≤ 0.05) and biological plausibility. Potential confounders included age, sex and prior TB history. For analyses restricted to those with HIV, we also evaluated the CD4 count and use of ART at baseline as potential confounders. Factors indicating severity of disease included semi-quantitative bacillary load (by AFB microscopy: scanty, 1+, 2+ and 3+) and chest X ray results. Potential confounders were considered actual confounders if their inclusion in the multivariable model changed the unadjusted odds ratio (OR) by 10% or more. In addition, given that our main measures of TB exposure were highly collinear, different models were built for each of those variables. Finally, in an attempt to distinguish patients with the highest probability of having a relapse of their prior TB disease from those who develop active disease due to a nosocomial exposure, we performed separate analyses for the total population and the population of patients who were not diagnosed with TB during their hospital admission. Two-sided p-value <0.05 was considered statistically significant. We used STATA/IC version 13.0 (StataCorp Inc, College Station, Texas) for the analyses.

The Human Research Committee of the Government of Botswana and the hospital’s Ethics Committee approved this study. The need for informed consent from the participant was waived by both Research Ethic Review Committees.

1,283 patients were scheduled for admission to the medical ward. However, 189 (6.8%) of them died at the Emergency Department before they could be transferred to the wards. 34 (18%) of these 189 patients had the working diagnosis of pneumonia. TB was retrospectively confirmed through positive AFBs or culture in 4 (12%) and remained as the most probable ethologic agent on 16 (47%) of those 34 cases of pneumonia (Additional file 2).

1094 patients were admitted to the medical wards during the study period. Demographic, clinical and radiographic characteristics of these patients are shown in Tables 1 and 2. Additional file 2 shows the distribution of patients according to the place of admission (cohorted vs. not cohorted) and diagnosis of PTB. Of the admitted patients, 131 (12%) spent their entire hospital stay at the TB cohorting bay and an additional 80 (7.3%) spent part of their stay in the TB cohorting bay (Additional file 2). The mean number of days of hospitalization was 7.8 (standard deviation [SD] 6.6) for the entire population, 7.3 (5.4) for HIV-uninfected and 8.2 (7.2) for HIV-infected (Table 3). During the hospitalization period, 198 (18%) died. PTB was diagnosed in 164 of the 896 patients (18%) who were eventually discharged.

77/896 (8.6%) patients were considered lost-to-follow up (or untraceable). During the 1-year follow up period, 41/896 patients developed confirmed PTB. The overall (all-cause) one-year mortality during follow up was 14% (123/896 patients). Of these, 19/123 (15%) had microbiologically proven TB and 7/123 (6%) had clinically diagnosed TB (Table 3). 72/896 patients were initially admitted to the general medical ward without a suspicion for PTB (Additional file 2).

One-year TB incidence and one-year TB-associated mortality were associated with admission in the cohorting bay, HIV infection, CD4 < 100 cells/mL, and PTB diagnosis during admission (Table 3). Bivariate analysis is shown in the Additional file 3. In multivariable analysis, one-year TB incidence was independently associated with the number of days that the patient remained hospitalized and proximity to TB index cases (exposure index) among patients without HIV infection (Table 4). Among HIV-infected patients, one-year TB incidence was independently associated with the number of days that the patient remained hospitalized, but the association with the exposure index was only found among HIV-infected patients who did not have a PTB diagnosis during admission. As expected, history of prior TB infection (particularly resistant TB), a diagnosis of PTB during hospitalization and older age were also associated with TB incidence during follow up (Table 4).