This large cohort study of 498 patients initiating ART in Durban, South Africa provides the most comprehensive evidence to date on the frequency, clinical spectrum, timing, outcome, and programmatic impact of IRIS relative to other clinical events following ART initiation in a sub-Saharan African setting. Particular strengths of our study include the prospective data collection, rigorous ascertainment and assessment of all clinical events, a standardised approach to defining IRIS with inclusion of cases at the milder end of the spectrum of IRIS, and differentiation between paradoxical and unmasking IRIS events and between OIs and less serious conditions.
In this study, the cumulative incidence of IRIS was 22.9%, or 65 cases per 100 PY, of which 64% were unmasking and 36% paradoxical. The incidence rates of the different IRIS subtypes were: paradoxical major OI, 13.3% of patients with an existing OI at the time of ART initiation; unmasking major OI, 5.0% of the entire cohort; paradoxical mucocutaneous, 9.3% of patients with relevant pre-existing conditions; and unmasking mucocutaneous, 10.4% of the entire cohort. One quarter of all IRIS events were due to TB, with other OIs accounting for only 10 cases (7%). The remaining 68% of IRIS events were due to mucocutaneous conditions, mainly immune reconstitution folliculitis, a well-recognised but poorly quantified IRIS phenomenon 
, followed by viral skin conditions such as herpes simplex, warts and herpes zoster.
Our overall IRIS rate and clinical spectrum is comparable to a London cohort of predominantly black African patients 
, which reported 22.7% cumulative incidence (20% due to major OIs or HBV, and 80% to dermatological conditions), but higher than the 10–17% reported from studies in other RLS 
. This may be partly explained by the low ascertainment of mucocutaneous IRIS events in these studies, lower rates of paradoxical TB-IRIS 
and possible missed ascertainment of early IRIS events due to death or loss to follow-up (median IRIS onset 48 days compared to 15 days in our study). Our observed 4.8% incidence of unmasking TB-IRIS and combined 0.8% incidence of unmasking and paradoxical cryptococcal IRIS were similar to those reported in studies from other RLS for TB (3.7–5.9%) 
and cryptococcal disease (0.3–1.6%) 
. The incidence of mucocutaneous IRIS (19.1%) was more consistent with results from retrospective studies conducted in higher income settings (9–20% 
) than in RLS (2–8% 
The wide range of incidence of IRIS reported across studies may be explained by variations in patient-, site- and setting-specific factors 
, quality of screening prior to ART, or by case ascertainment and definition methods. Comparisons across studies are further constrained by the failure in most studies to discriminate between paradoxical and unmasking presentations, and the lack of a standardised case definition for IRIS. In the absence of clear immunopathological correlates or diagnostic assays, the differentiation of IRIS from other competing diagnoses and particularly unmasking IRIS from a new infection, or paradoxical IRIS from natural history of pre-existing disease, remains challenging. While no single case definition can comprehensively address all possible presentations of IRIS, the progressive development and use of both generic 
and pathogen-specific 
case definitions in this and future studies of IRIS will improve reliability and comparability across studies.
We also evaluated the relative importance of IRIS compared to early non-IRIS clinical events. IRIS accounted for almost one quarter of all clinical events and deaths and one fifth of hospitalisations in the initial 24 weeks after ART initiation, which was similar to the contribution of non-IRIS new infections. Of the 114 patients who developed any type of IRIS event, 7% died, similar to the IRIS mortality of 5% and 9% reported in two studies from RLS 
. However, among the subgroup of patients with IRIS related to major OIs, there was a much higher morbidity and mortality (16.3% died, 34.9% were hospitalised, and 4.7% discontinued ART). Overall, there was no apparent negative impact of IRIS on ART adherence and retention, although follow-up was more intensive in our study than would be usual in most settings. Drug toxicity accounted for few events or hospitalisations, and no deaths, but almost half of ART discontinuations and regimen changes.
The baseline characteristics of our study population (predominantly young, female, with a high prevalence of WHO stage 3–4 disease and very low baseline CD4+
counts) are typical of other high HIV prevalence settings in sub-Saharan Africa 
. We therefore extrapolated our findings to estimate the wider programmatic impact of IRIS. By 2009, approximately 972,000 patients had initiated ART in South Africa 
. Assuming a 5% incidence of unmasking OI-IRIS, a 20% prevalence of co-infection with major OIs at ART initiation, a 13% incidence of paradoxical IRIS, and case fatality and hospitalisation rates of major OI-IRIS of 13.6% and 27.3%, we estimate that IRIS may have caused around 10,000 deaths and 20,000 hospital admissions in the South African ART programme alone since inception. If we apply the same assumptions to the 2.2 million individuals who have initiated ART across the ten countries with the largest ART programmes, IRIS may have accounted for as many as 23,000 deaths in these countries. The impact of IRIS is likely to increase as ART programmes in Africa are integrated with inpatient services, and ART is increasingly offered to those with major OIs and more advanced disease. However, it is important to emphasise that the effects of IRIS were similar to the number of events, deaths and hospitalisations caused by new non-IRIS infections, but slightly less than those caused by pre-existing conditions.
We identified some important differences in risk factor profiles between paradoxical and unmasking, and major OI-associated and mucocutaneous IRIS, consistent with these being distinct entities with differing pathogenesis, although these subgroup analyses were limited by small numbers. Consistent with previous studies, lower CD4+
and higher VL 
at baseline were predictors of paradoxical OI (mainly TB)-IRIS, but not unmasking IRIS, together with shorter period of pre-ART duration of OI/TB therapy OI-IRIS 
, which is likely to be related to a higher residual pathogen load. In contrast with several published studies, the magnitude and rate of CD4+
immune reconstitution 
and VL reduction following ART initiation 
were not predictive of any type of IRIS. However, comparison across studies is difficult because of the potential confounding effect of a low baseline CD4+
count, and the use of CD4+
and VL responses in defining IRIS in some studies 
For unmasking OI (predominantly TB) IRIS, the risk factors we identified were all consistent with the presence of sub-clinical disease, in particular lymphadenopathy present on chest X-ray, low haemoglobin, greater than 10% weight loss, and high baseline CRP. All were associated with at least a two- to three-fold increased risk of IRIS, and in the case of lymphadenopathy, a nine-fold risk. There are few existing data on risk factors for unmasking IRIS. One recent small study from Uganda of unmasking TB-IRIS identified both a low BMI and high CRP as independent predictors 
, but an Indian study identified no specific risk factors 
. This is the first study to examine risk factors specific for mucocutaneous IRIS – the most common type of IRIS. Not surprisingly, with such a diverse range of conditions encompassing viral and bacterial aetiologies, identification of common risk factors was less productive. A low baseline CD4+
count was weakly predictive of paradoxical MC IRIS, and mild anaemia and a history of current or past TB were associated with unmasking MC IRIS.
There are several practical implications of our findings for reducing IRIS-associated morbidity and mortality in ART programmes. First, earlier HIV diagnosis and initiation of ART in accordance with revised WHO guidelines at a CD4+ count of 350 cells, and certainly before a decline to below 200 cells, remains the most important strategy to reduce the prevalence of major OIs, and the incidence of most serious and life-threatening IRIS and non-IRIS events. We estimate that for each 50 cell increase in baseline CD4 count at ART initiation, there would be a 17% reduction in the risk of any IRIS (95% CI 6–26%), and a 43% reduction in all-cause mortality (95% CI 19–60%).
Second, to reduce the likelihood of paradoxical OI IRIS, there is a need to optimise pre-ART OI management and timing of ART. Consistent with other reports, we found that a shorter duration of pre-ART OI/TB therapy was predictive of OI-IRIS 
, which is likely to be related to a higher residual pathogen load. The optimal timing of ART in patients with OIs requires balancing greater risk of IRIS after early initiation against continuing high mortality with delayed ART. Several RCTs of TB show a mortality benefit from earlier ART in TB 
, while one RCT suggested a benefit of deferral for ten weeks with cryptococcal infection 
. While there are limitations in using observational data to inform this debate, our data can be viewed in conjunction with these RCTs and other IRIS cohort studies.
Third, there is a need to improve screening for OIs, particularly TB and cryptococcus, prior to ART initiation, to reduce the development of unmasking IRIS of major OIs, which are associated with a particularly poor outcome 
. Several features suggestive of sub-clinical TB were predictive for unmasking TB in our study, suggesting that existing practices for pre-ART screening and treatment are inadequate. HIV programmes in high TB prevalence settings should incorporate a simple validated symptom score 
(with additional chest x-ray and sputum examination as required) into routine pre-ART evaluation, as recommended in WHO 2010 guidelines on intensified case finding 
. This will be further enhanced by the increasing availability of rapid point of care tests for TB diagnosis 
. Screening for cryptococcal antigen prior to ART initiation in patients with a CD4+
count less than 100 cells with pre-emptive fluconazole treatment in those who are serum CrAg positive has been shown to be a cost-effective intervention to reduce mortality 
and this is now a recommendation for high cryptococcal prevalence settings in recent WHO advice 
. Further operational research is needed to determine the feasibility and impact of a serum CrAg “screen and treat” strategy 
More intensive monitoring of selected patients at high risk of IRIS (e.g. CD4+ count <50 cells, known OI, unexplained symptoms) in the initial few weeks after starting ART is a further strategy to reduce IRIS-related mortality by facilitating prompt management of severe IRIS. IRIS is generally self-limiting and interruption of ART is rarely indicated, but patients may require reassurance in the face of protracted symptoms to prevent discontinuation of or poor adherence to ART. Few national HIV guidelines from RLS contain guidance on recognition and management of IRIS. There is a need for both better health care worker training in this area as well as improved patient awareness.