Introduction

Tuberculosis (TB) continues to be a major public health issue worldwide, particularly in low and middle-income countries despite rigorous efforts to contain its spread and implementation of effective treatment strategies. In 2014 an estimated 12 million people worldwide were living with active pulmonary TB, with 9.6 million new cases and 1.5 million deaths due to TB occurring annually1,2,3,4,5,6,7.

TB does not usually require hospital admission for treatment, but if symptoms such as shortness of breath, and deterioration in a systemic condition are present, hospitalization may be necessary. A large proportion of patients with TB are hospitalized8,9, and estimates of in-hospital mortality range from 2% to 12%10,11,12,13,14; most of the current costs of TB treatment result from hospitalization15.

A variety of predictors have been associated with a greater risk of death among TB patients, including poverty, homelessness, alcohol or drug addiction, irregular or inadequate treatment, late diagnosis of the disease, multidrug-resistant TB (MDR-TB), and advanced age4,6. Human immunodeficiency virus (HIV) infection is an important factor related to the increased morbidity and mortality of TB in different world regions4,10. In addition, diabetes has been reported to be associated with increased risk of mortality16,17,18. Also, men have higher rates of mortality and worse outcomes compared with women19,20. Previous TB with multiple treatments has also been associated with in-hospital mortality21,22,23. Furthermore, patients with malignant tumors are immunocompromised and can have unusual clinical presentations, both related to delayed diagnosis and high mortality24,25,26.

In TB program monitoring, TB deaths are crucial indicators of the impact of TB control measures10,11,12,13,14, especially in areas with high HIV and TB prevalence. Data on TB deaths should provide us with a better understanding of the factors associated with these deaths and help guide interventions to reduce mortality; however, there is uncertainty regarding which factors are associated with in-hospital mortality among patients with pulmonary TB10.

We therefore conducted a systematic review and meta-analysis to establish predictors of in-hospital mortality among patients with pulmonary TB.

Methods

Search strategy

We used a multimodal search strategy focused on 3 bibliographical databases (MEDLINE, EMBASE and Global Health). An experienced librarian (RC) used medical subject headings, adding terms and keywords from a preliminary search to develop the database search strategies. In each database, the librarian used an iterative process to refine the search strategy through testing several search terms and incorporating new search terms as new relevant citations were identified. There were no language restrictions. The search included the following databases from inception to November 2015: MEDLINE, EMBASE and Global Health. The search consisted of three concepts combined using the AND operator1: tuberculosis2, hospitalization and3 mortality (Appendix 1). The protocol of this study was published elsewhere27.

Study selection

Eligibility criteria

Eligible trials met the following criteria1: cohort or case-control design2; explored risk factors for in-hospital mortality among patients with pulmonary TB in an adjusted analysis.

Assessment of study eligibility

Two reviewers (CPBA and DRS) trained in health research methodology screened, independently and in duplicate, the titles and abstracts of all citations identified in our search. The same reviewers screened all full text articles for eligibility; disagreements were resolved by consensus, with consultation of a third investigator (JWB) when resolution could not be achieved. We measured agreement between reviewers with the kappa statistic to assess the reliability of full-text review using the guidelines proposed by Landis and Koch28: <0.20 as slight agreement, 0.21–0.40 as fair agreement, 0.41–0.60 as moderate agreement, 0.61–0.80 as substantial agreement and >0.80 as almost perfect agreement.

Assessment of study quality

Two reviewers (CPBA and DRS) assessed risk of bias for each eligible study, independently and in duplicate, using the Newcastle-Ottawa quality assessment scale (NOS) for Cohort Studies29. The scale consists of nine items that cover three dimensions1: patient selection (4 items)2; comparability of cohorts on the basis of the design or analysis (2 items); and3 assessment of outcome (3 items). A point is awarded for each item that is satisfied by the study. The total score therefore ranges from zero to nine, with higher scores indicating higher quality. A total score ≥7 represents high quality.

Data Extraction and Analysis

Two reviewers (CPBA and DRS) extracted data from each eligible study, including demographic information (e.g. sex, age, race), methodology, and all reported predictors.

We performed meta-analysis for all predictors that were reported by more than one study. We used odds ratios (ORs) with associated 95% CI to measure the association of binary predictors and in-hospital mortality. We used random effects models for all meta-analyses. If a study reported more than 1 regression model, we used data from the most fully adjusted model presented. We also presented the results from the predictors explored by the studies but that were not eligible for meta-analysis.

We evaluated heterogeneity for all pooled estimates through visual inspection of forest plots, because statistical tests of heterogeneity can be misleading when sample sizes are large and CIs are therefore narrow30. We used the software R.

Publication bias

For meta-analyses with at least 10 studies, we assessed publication bias by visual assessment of asymmetry of the funnel plot and performed the Begg rank correlation test31.

Quality of evidence

We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to summarize the quality of evidence for all meta-analyses32. We categorized the confidence in estimates (quality of evidence) as high, moderate, low or very low, on the basis of risk of bias33, imprecision34, indirectness, inconsistency35 and publication bias36. We used GRADE evidence profiles to provide a succinct, easily digestible presentation of the quality of evidence and magnitude of associations32. In case of doubt or missing details about the studies, authors were contacted for clarification.

Ethics and Dissemination

This study is based on published data, and therefore ethical approval was not a requirement. This systematic review and meta-analysis is expected to serve as a basis for evidence to reduce in-hospital mortality in TB patients, and as a guide for future research based on identified knowledge gaps. It is anticipated that findings from this review will be useful for informing policy, practice and research priorities, improving the management of in-hospital TB patients. We also plan to update the review in the future to monitor changes and guide health services and policy solutions.

Results

Search Results and Study Characteristics

We identified 2,969 unique records, of which we retrieved 51 English and 3 non-English language articles in full text; 11 cohort studies, published between 2003 and 2013, that evaluated 5,468 patients proved eligible. Figure 1 shows the study selection flow diagram. There was substantial agreement (κ = 0.64) at the titles and abstract screening stage and perfect agreement (κ = 1.00) between reviewers at the full-text review stage.

Figure 1
figure 1

Flow diagram of study selection.

Full size image

All 11 eligible studies1,4,15,37,38,39,40,41,42,43,44 were single-center and there was one non-English (Chinese) study included in our analysis. Two studies38,42 were conducted in Japan, two40,41 in Taiwan, three15,39,43 in Korea, one37 in Germany, one4 in Israel, one1 in Iran and one44 in China. One study39 used TB-related mortality as defined by the World Health Organization (the number of TB patients who died during treatment, irrespective of cause)45, two38,42 used all-cause mortality, and eight1,4,15,37,40,41,43,44 used TB-related mortality as judged by the investigators. The majority (9 of 11)1,4,15,37,39,40,41,43,44 acquired data from medical records, with eight retrospective cohorts1,4,37,38,39,40,41,42 and one prospective cohort study15 (Table 1).

Table 1 Studies describing in-hospital mortality among pulmonary tuberculosis patients.
Full size table

Risk of bias

Overall, the quality, evaluated by the NOS checklist for the outcome “mortality”, was high (Table 2). We did not have a sufficient number of studies in our meta-analyses to assess publication bias.

Table 2 Newcastle-Ottawa scoring system for cohort studies.
Full size table

Predictors of in-hospital mortality

A total of 11 studies, involving a total of 2343 patients, reported the association of 60 factors with in-hospital mortality1,4,15,37,38,39,40,41,42. On the basis of our criteria, we conducted meta-analyses for 5 predictors of in-hospital mortality1: acid-fast bacilli (AFB) smear positive2, diabetes mellitus3, malignancy4, history of previous TB, and5 male sex.

Moderate quality evidence showed a significant association between malignancy and in-hospital mortality among TB patients (OR 1.85; 95% CI 1.01–3.40). Low quality evidence showed no association between in-hospital mortality and AFB smear positive test (OR 0.99; 95% CI 0.40–2.48), or male sex (OR 1.09; 95% CI 0.84–1.41). Very low quality evidence showed no association between mortality and diabetes mellitus (OR 1.31; 95% CI 0.38–4.46), or previous TB (OR 2.66; 95% CI: 0.48–14.87) (Fig. 2; Table 3).

Figure 2
figure 2

Association between AFB smear positive, Diabetes Mellitus, Hx of previous TB, Malignancy, male sex and in-hospital mortality among pulmonary TB patients.

Full size image
Table 3 GRADE Evidence Profile: Predictors of in-hospital mortality among TB patients.
Full size table

Table 4 presents the associations with in-hospital mortality for the factors that were not amenable to meta-analysis.

Table 4 Unpooled predictors for in-hospital mortality among TB patients.
Full size table

Discussion

We found moderate quality evidence that co-morbid malignancy was associated with increased in-hospital mortality among TB patients. Low quality evidence showed that sex and AFB smear positive were not associated with in-hospital mortality, and very low quality evidence showed no association with previous TB infection and diabetes mellitus.

Our review has a number of strengths. Our search, which had no language restrictions, was designed and implemented by a research librarian, and literature screening and data extraction were performed independently and in duplicate by two reviewers using pretested, standardized extraction forms. The main limitation of our review was the small numbers of events that contributed to our meta-analyses, resulting in wide estimates of precision for our pooled measures of association.

Other studies24,25,26 also found that malignancy increases the risk of death in TB patients. Patients with malignant tumors are immunocompromised due to the local or systemic effects of the disease itself, as well as to the treatment regimens, which can impair the immune system and make these patients particularly susceptible to developing TB46. In addition, TB can have an unusual clinical presentation, making diagnosis more difficult in these patients, contributing to delay in diagnosis and high mortality rates47,48.

While not significantly associated with mortality in our review, previous TB has been reported to be associated with in-hospital mortality in many studies1,21,22,23. Patients who undergo multiple treatment regimens for TB can develop resistance to drugs with the subsequent emergence of MDR-TB and XDR-TB, conditions highly associated with greater risk of death21. Further, in settings other than hospitals, studies49,50 have demonstrated that smear positive patients have a better prognosis regarding mortality than smear negative patients. Indeed, indicators of atypical manifestations, such as smear-negative sputum, were associated with delayed diagnosis and mortality12,51. Recently, a retrospective cohort study from Brazil6 reported a high mortality rate during hospitalization (16.1%), and negative sputum smear microscopy was an in-hospital mortality predictor in the population studied. However, patients with pulmonary and extrapulmonary TB were included in this study.

We did not find a significant association between male sex and in-hospital mortality among pulmonary TB patients. Worldwide TB notification data show that far more men than women have TB7. Some studies showed that mortality rates are higher in females during their reproductive years, but after that they are higher in men19,20.

Diabetes was also not associated with mortality in pulmonary TB patients in this study. Only one study1 included in this meta-analysis showed that diabetes was a predictor of mortality in TB patients, possibly because they included a larger number of diabetes patients (18% of the enrolled individuals). Some studies1,16,17,18 have found that diabetes increases risk of early mortality during TB treatment. This effect may be explained by impaired TB treatment response16.

In conclusion, the presence of malignancy was significantly associated with in-hospital death in pulmonary TB patients. Other predictors were not associated with in-hospital mortality in TB patients, probably due to the small number of events. Further research should explore promising predictors of in-hospital mortality in large prospective studies.