Our study demonstrated that over two-thirds of hospitalised HIV-TB patients develop complications resulting in clinical deterioration after starting ART as inpatients. Over 40% experienced paradoxical TB-IRIS, with HAIs, drug toxicities, and opportunistic diseases being other frequent causes of deterioration. A novel finding is that HAIs were common (occurred in 20.5%), and 11 of the 12 deaths were attributed to sepsis. Reassuringly, despite the high incidence of clinical deterioration, 89% of participants were alive three months after ART initiation. Adequate clinical care and diagnostic and therapeutic resources are required to manage this HIV- as well as tuberculosis-associated morbidity.
HIV infection, especially with a low CD4 count, is a risk factor for clinical deterioration and death on TB treatment. 
A previous study conducted in Cape Town of 292 ambulatory TB patients, 209 of whom were HIV-infected, demonstrated that 40% deteriorated while on TB therapy, 26% required hospitalization, and 15% died during the 6 month follow-up period. 
Training health care workers on the causes, recognition and management of deterioration in HIV-TB patients, in addition to training on ART management, is required to optimize the outcome of HIV-TB patients.
The high incidence of paradoxical TB-IRIS in our study (42%) corresponds with the higher estimates reported from previous cohort studies (8–43%) 
. Our patients had many of the risk factors for developing TB-IRIS; low CD4 counts, disseminated tuberculosis and relatively short time interval between starting tuberculosis therapy and ART. Our findings add to the evidence that paradoxical TB-IRIS is an uncommon cause of mortality 
, even in very ill HIV-TB patients starting ART in hospital. One caveat is that 7 of 47 IRIS cases developed potentially life-threatening manifestations and these were promptly ascertained and managed by specialists with early corticosteroid introduction or dose escalation. The association of a raised CRP prior to starting ART with paradoxical TB-IRIS has been previously described. 
Being on corticosteroids at the time of initiating ART showed a trend towards reduced risk for TB-IRIS, although this was not statistically significant in this small study. Larger prospective studies would be necessary to evaluate this properly.
The large number of HAIs we diagnosed, and the associated mortality is alarming. Our findings suggest that once patients with advanced HIV and TB are established on appropriate TB treatment, ART and cotrimoxazole, then bacterial infections emerge as the most important cause of death. We suspect that colonization with multi-drug resistant bacteria occurred in the referring general hospital, as antibiotic selection pressure in TB hospitals is likely to be low. High rates of urinary catheterization during the acute admission at the referring hospital, is a likely contributing risk factor to the urinary tract infections. Depressed monocyte responses may be one of the reasons why HIV-infected patients with advanced immunosuppression are at high risk of bacterial infections. 
Our study and a previous study from a hospitalized cohort in Cape Town 
, confirm that drug-resistant bacteria which require the use of carbapenems and other costly antibiotics are causing HAI. These antibiotics are generally unavailable at district or secondary level hospitals in developing countries and there is therefore often a delay in treating these infections appropriately resulting in high mortality. Furthermore, because the risk of multi-drug resistance is high in HAIs, it is essential to culture clinical specimens prior to commencement of antibiotics. Our findings emphasize not only the need for appropriate antibiotics to be available, but for basic infection prevention control practices, most importantly effective hand disinfection, to be re-inforced and practiced to prevent secondary spread of infection. Reducing the number of days that patients have urinary catheters in situ, and wherever possible avoiding the need for indwelling intravenous catheters are also important interventions. Every attempt is made to limit duration of hospital stay. However, severity of illness and extremely poor social circumstances often preclude early discharge.
Our study has several limitations. The follow-up period was limited to the first 3 months of ART, so events occurring in the latter half of TB therapy were not ascertained. The generalizability of our findings is limited by the fact that all participants in the study were evaluated by a specialist physician trained in infectious diseases at study visits, which could in part explain our relatively low mortality. There was no control group to ascertain outcome of patients who did not receive input from an Infectious Diseases specialist. Another factor limiting generalizability of our outcomes is that BCH is relatively well-resourced with access to radiology and laboratory services. It is unlikely that such good outcomes can be achieved in less well-resourced facilities, which would be found in the majority of TB hospitals in low- to middle-income countries. However, specialist input and access to diagnostic facilities allowed us to ascertain the frequency and type of complications occurring in sick HIV-TB inpatients starting ART. Another limitation is that not all patients developing sepsis had appropriate cultures sent, as transportation of specimens to the off-site microbiology laboratory occurs only once a day during weekdays. In addition the causes of death were ascertained by the attending infectious diseases specialist and no post-mortem studies were performed on the patients. For the analysis of predictors of mortality the sample size was to too small to allow for meaningful comparison between groups.
Our findings have important implications for service delivery and resource allocation at hospitals offering dedicated tuberculosis services. HIV-TB patients requiring long term admission are a vulnerable group with advanced immunosuppression, frequently have disseminated tuberculosis, and are at high risk of early mortality if ART is not initiated timeously. Three landmark trials have shown that ART initiated within 2–3 weeks after starting TB therapy in patients with HIV-associated tuberculosis and a CD4<50 cells/mm3
results in a reduction in death and AIDS.
Where HIV-TB services are not integrated, hospitalized HIV-TB patients are referred to primary community ART clinics after discharge, resulting in unnecessary delay of ART initiation and this likely increases mortality. If HIV-TB patients have to attend outpatient ART clinics to initiate ART during their hospital stay, there is a risk of transmitting M. tuberculosis
, including drug-resistant strains to the HIV-infected clinic population. Hence, integrating ART initiation into tuberculosis hospitals is vital to expedite ART initiation as well as to prevent TB transmission in ART clinics. With proper human resource allocation to tuberculosis hospitals such as BCH, ‘fast-track’ counselling can be employed to ensure patients are treatment ready. Patients in our study started ART a median of 36 days after TB treatment, much of the delay occurring because of the period it took for patients to be admitted to BCH after TB diagnosis.
The strength of our study is that it illustrates the clinical complexity of starting ART in hospitalized HIV-TB inpatients and the need for clinicians trained in the clinical management of these very ill patients. It argues strongly for proper training and resource allocation to manage this vulnerable group of patients.
High rates of paradoxical TB-IRIS, HAIs, drug toxicities and new opportunistic infections occur in hospitalised HIV-TB inpatients with advanced immunosuppression initiating ART. Despite the high morbidity, relatively good short-term outcomes can be achieved with careful clinical management. There is a need for prospective studies in other settings, and for additional studies evaluating the effect of training and the provision of ART to HIV-TB inpatients.