All five patients discussed above had a diagnosis of tuberculosis made pre-HAART and were on TB therapy at time of HAART initiation. They were immunosuppressed, as evidenced by clinical findings and CD4 count. All presented with new tender hepatomegaly, jaundice with bile canalicular enzyme rise and preserved liver synthetic function, as evidenced by normal clotting studies and return of constitutional symptoms within 8 weeks of initiation of HAART. Ultrasound scan was not diagnostic in any of the patients; in particular none had bile duct dilatation. There was evidence of multiple potential aetiologies in all patients, but the predominant cause was felt to be DILI in one case and hepatic TB- IRIS in the other four. We discuss these findings in the context of previous literature under the headings liver disease in HIV infection, drug induced liver injury and hepatic TB IRIS below.
Liver disease in HIV infection
In the pre-HAART era abnormal liver enzymes were regarded common in HIV, with prevalence rising with advancing immunosuppression [4, 5]. Histopathology studies demonstrated a multiplicity of HIV associated disease in the liver, with aetiology differing widely in different population settings [6–10]. Despite this, jaundice has been regarded as infrequent in HIV . Again, as might be expected, the incidence and aetiology of jaundice in HIV-infected patients may vary in different populations [12, 13].
Since the introduction of HAART there has understandably been greater focus on drug associated hepatotoxicity in HIV. In high income settings, chronic liver disease – associated with Hepatitis C and B virus, long term drug-induced toxicity, alcohol related and non-alcoholic fatty liver disease – has become a leading cause of chronic morbidity and mortality in people living with HIV . Mechanisms underlying long-term hepatic injury are probably distinct from hepatotoxicity seen acutely after introduction of HAART . Early severe hepatotoxicity (SH) is often arbitrarily defined as elevated serum aminotransferases >5 times the upper limit of normal (grade 3 and above by the AIDS Clinical Trial Group classification) within the first 6 months of HAART; in randomised controlled trials (RCTs) incidence of SH ranges from 2 to 18% , and may have even greater variability in observational studies .
However there are well described problems with assessing hepatotoxicity of HAART in randomised and observational trials . Asymptomatic elevations in liver enzymes may not be an accurate surrogate for risk of developing rare, clinically relevant, liver disease. Furthermore, the biochemical definition of hepatotoxicity is not standardised despite use of the AIDS Clinical trial Group (ACTG) classification. Although ACTG give parameters for significant elevation of all liver enzymes , many studies only include transaminase elevations – raised total bilirubin is often inconsistently defined, and alkaline phosphatase rarely considered. Which liver enzymes are included can have a large effect on conclusions . In addition, when considering patients with baseline abnormalities of transaminases investigators generally use modified ACTG criteria (for example elevations of transaminases > 3.5 times baseline), but such modifications are not standardised. Finally, upsets in liver enzymes can occur in the absence of HAART, but in hepatotoxicity studies all significant hepatic abnormalities are attributed to “toxicity”, or it is often not clear how investigators have excluded cases due to other causes.
Nevertheless, it is clear that, again, in high income settings co-infection with viral hepatitides is pivotal in determining risk of early hepatotoxicity [21–24]. By comparison, HIV cohorts in limited resource settings tend to have much lower rates of Hepatitis C infection, and are more likely to use nevirapine in first line regimens. Hepatotoxicity studies in these settings reveal a heterogeneous picture. Reported rates of early hepatotoxicity in people initiating HAART in African studies are generally low at 1-2% [25–33] but occasionally much higher at 3.4-18.4% [34–37]. Grade 1 or 2 rises in liver enzymes are very prevalent in all these studies. Some well conducted studies have reported a significantly higher rate of early grade 3 or 4 hepatotoxicity in patients treated with nevirapine compared to efavarenz [25, 37], while other studies have not . Co-infection with viral hepatitides is not found to be a significant risk factor for hepatotoxicity [25, 26, 28, 34, 37], in most but not all studies . In contrast, co-treatment with anti-tuberculous medications is found to be a major risk factor in African cohorts [25, 28, 35], although this also has exceptions .
Different follow up times and frequencies of serum liver enzyme assessment may explain much of this variation – for example the studies which have found high rates of early nevirapine associated hepatotoxicity have higher rates of blood sampling in the first 3 months post-HAART initiation, which is when most hepatotoxicity was identified, suggesting other studies may simply have missed this occurrence.
In correspondence with the high income setting studies, there is also the problem of definition of hepatotoxicity. This difficulty is underlined in light of the finding that point prevalence of ‘grade 3 or 4 hepatotoxicity’ may be very similar before and after HAART initiation , or that HAART initiation even reduces the rates of transaminitis in African HIV treatment cohorts . This suggests trials which define HAART related hepatotoxicity as any significant rise in liver enzymes after HAART initiation have oversimplified a complex, multifactorial problem.
Helping get beyond these problems are papers that report clinically relevant outcomes, which several studies in an African setting do. Reported rate of changing HAART regimen in response to hepatotoxicity is gene rally lower than reported rate of grade 3 or 4 liver enzyme rise – for example 0.7% v. 1.4%;  9.1% v. 13.8%;; 2.6% v. 3.4%;  0.9% v. 4.6% . It is possible that clinicians do not always think such rises are heralding life threatening drug induced liver injury, perhaps relying instead on additional features such as lactic acidosis, hypersensitivity rash or eosinophilia or are reassured by the transient nature of a rise. Several investigators have recorded that grade 3 or 4 liver enzyme abnormalities often resolve spontaneously without change in medications [29, 37]. However, some investigators have noted higher incidence of symptoms in those with compared to those without grade 3 or 4 liver enzyme elevation [35, 37]. Importantly, while deaths attributed to HAART hepatotoxicity are infrequent, they do occur; reported rates are 0 to 0.5% [35–37].
Further clinical context is provided in some hepatotoxicity studies. In reports of hepatotoxicity in a treatment cohort in rural Uganda [28, 38], Weidle and colleagues describe clinical hepatitis – defined as presence of jaundice, liver enlargement and gastrointestinal symptoms – in 4 out of 1029 enrolled patients within 3 months of HAART initiation. Although 12.7% of this cohort received TB treatment, none of the 4 patients with clinical hepatitis were on anti-mycobacterial medication at the time of their reactions. All 4 patients also exhibited a hypersensitivity rash, one case was caused by co-trimoxazole, and 3 were attributed to nevirapine; one of the 4 patients died from the reaction.
Kalyesubula et al. reporting from urban clinics in Kampala give a detailed account of clinical and biochemical surveillance for hepatitis in HIV seropositive patients . In a cohort of 236 individuals, 66 developed new transaminitis within 14 weeks of commencing HAART, although only 3/66 were grade 3 or 4 rises. The investigators report that, in the 66 patients with any aminotransferase elevation, 33% had vomiting, 20% right upper quadrant pain, 17% hepatomegaly and 8% jaundice. Although only 8/236 patients were receiving any TB treatment, the authors found current TB therapy to be a risk factor for any grade 2 – 4 transaminitis, but it is not clear what clinical findings were observed in HIV-TB co-infected patients who developed transaminitis.
Unlike in this previous work, the current case series is defined by a clinical presentation rather than the presence or absence liver enzyme elevation. In addition, we present a detailed clinico-pathological description in an attempt to better define the causes of the liver disease in these patients, which is not found in the literature on hepatotoxicity post-HAART. It should be emphasised that the treatment cohort these cases are drawn from may be quite atypical compared to some of the above studies. In particular, the very high rate of culture confirmed tuberculosis in our cohort (about 40% from samples obtained during HAART initiation work up).
Drug Induced Liver Injury (DILI)
There is no gold standard for the diagnosis of DILI. The clinical and histopathological spectrum of DILI is wide and overlaps with other hepatic diseases. The more common hepatitic pattern of DILI, with ALT >5 times upper limit of normal, was not apparent in these five patients who had predominant bile canalicular enzyme elevation. However, cholestatic or mixed hepatocellular/cholestatic DILI has been reported with use of rifampicin,  isoniazid , ethambutol , and co-trimoxazole , hence, drug hepatotoxicity must still be considered.
In lieu of a gold standard diagnostic test, clinical scales for diagnosis of DILI have been developed. One of the most validated is the Maria &Victorino (M&V) system . Based on the M&V system score, patient 5 above might be considered a ‘possible’ DILI, while patients 1 to 4 score as ‘unlikely’ to be DILI. Caution must be taken in application of DILI clinical scales: poor reproducibility along with inter-rater and inter-scale disagreement have been shown , and they are not validated in HIV- positive populations or low resource settings. Despite these limitations, from a clinical perspective, worsening of LFTs on drug re-challenge in patient 5 is strong evidence of DILI, and normalisation of LFTs without interrupting therapy is strong evidence that DILI was not a predominant cause of the jaundice seen in the four other cases.
Histopathologic assessment of possible DILI can also be problematic. Drug hepatotoxicity can produce a spectrum of acute hepatic pathology that imitates a range of patterns described in primary hepatic disease [45, 46]. This mimicry poses a major diagnostic challenge because the shared histopathological features, comprising mainly necro-inflammatory hepatocellular, cholestatic and mixed hepatocellular-cholestatic patterns, cannot be attributed unequivocally to drugs [45, 47]. Although this wide spectrum of histopathological reaction patterns displayed in DILI includes a granulomatous reaction pattern, drug-induced granulomatous responses are usually non-necrotising. Because 3/5 patients in the present study demonstrated necrotizing granulomatous inflammation, a drug-induced cause is morphologically inappropriate. This is also strengthened by the identification of acid fast bacilli in one biopsy and confirmation of an M. tuberculosis complex footprint by polymerase chain reaction.
One of the five biopsy results in these cases was considered strongly suggestive of DILI and resulted in decision to interrupt TB medications (case 5, Figure 3, histology showed moderate portal tract and lobular inflammation with eosinophils and interface hepatitis). The presence of eosinophils is a well-recognised feature of drug reactions in many organs, but caution is necessary in the interpretation of their presence. This is particularly pertinent in the current context, because patients with advanced AIDS demonstrate a dominant Th2 response with associated tissue “eosinophilia”. Furthermore, silent parasitic infections may be responsible for heightened eosinophil numbers. Notwithstanding this, however, whilst all biopsies in the present study contained eosinophils, the density of eosinophils was most intense in the biopsy from patient 5. In this biopsy, there were no other histopathological clues of a parasitic infestation.
In summary, no gold-standard exists for diagnosis of DILI, but liver biopsy with clinicopathological correlation suggests DILI as a probable primary cause of hepatic disease in only one of the five cases presented here.
Although tuberculosis of the liver is not commonly diagnosed clinically, it is a frequent post-mortem finding in HIV positive patients . It is possible that hepatic tuberculosis (IRIS) may be similarly under recognised. Lawn and Wood  report not infrequent involvement of the liver in South African patients with TB-IRIS – four patients in a case series of seventeen TB-IRIS patients. They suggest that predominant rise in bile canalicular enzymes, liver capsular pain, manifestations of TB-IRIS at another anatomical site, are all suggestive features, while jaundice may develop in some patients.
Meintjes et al.  have also shown high prevalence of cholestatic pattern liver enzyme derangement and hepatomegaly in South African patients diagnosed with TB-IRIS in a large prospective observational study, but do not comment specifically on jaundice. Similarly, Haddow et al.  report a high frequency of elevated liver enzymes in a cohort of South African patients initiating HAART who were consequently classified as possible cases of hepatic TB-IRIS by expert consensus. By contrast other TB-IRIS cohort studies do not report any liver involvement . Such variability may represent differences in study populations or may relate to diagnostic difficulty. In likeness with DILI, TB-IRIS has no diagnostic gold standard test, but a consensus clinical case definition has been proposed for low income settings . Four of the five jaundiced patients in this series were felt to have met this TB-IRIS case definition (Additional file 1: Table S2).
A specific case definition for hepatic TB-IRIS has also been suggested, which additionally requires the presence of granulomatous inflammation on liver biopsy . Lawn and Wood suggest that large epithelioid granulomatous inflammation seen on liver biopsy is evidence of hepatic TB-IRIS . In the present study, not only was necrotising and non-necrotising granulomatous inflammation present in the lobules and portal tracts, but the granulomas demonstrated – in addition to epithelioid histiocytes and Langhans giant cells – neutrophils, plasma cells and large numbers of lymphocytes, which are not features of a conventional untreated tuberculous response. The microscopic spectrum of the tuberculous IRIS reaction is poorly documented, but a similar dense neutrophilic and lymphoplasmacytic inflammatory response has been documented in reaction to cryptococcal antigen in tissue . Although the present study is limited by the absence of interval CD4 counts and viral load measurements, in view of the other laboratory findings and clinical history, including patient outcome, the spectrum of histopathological findings is proposed as the expanding morphological profile of tuberculous granulomas in the setting of IRIS. We believe this is the most detailed clinico-pathological description of hepatic TB-IRIS to date. Further multi-organ based histopathological studies in multiple organs are necessary for such validation.
Multiple causes of jaundice in individual patient
Immune-compromised patients are often found to have multiple concurrent liver pathologies . The majority of cases presented above have evidence of more than one potential cause of jaundice. A further contributor to jaundice in these cases could be the isolated hyperbilirubinaemia (without evidence of cholestasis or hepatocellular damage) related to rifampicin interference with bilirubin uptake [56–58]. Heller et al.  postulate this may be at least a partial cause of their ‘relatively frequent’ observation of jaundice shortly after initiation of TB medications in South African patients. This might be particularly relevant to patient 5 in this case series, who had jaundice in the presence of only minor alkaline phosphatase elevation.
Preliminary guidelines on how healthcare workers in resource-limited settings should respond to jaundice in HIV positive TB cases on antiretroviral therapy
A lack of evidence exists informing the management of the clinical presentation of jaundice in this context and the preliminary nature of the following guidance should be emphasised.
Presentation with jaundice in HIV positive patients on TB treatment and HAART should prompt full clinical assessment. Bloods for biochemical and haematology assays, and ultrasound assessment of the abdomen should be performed if available.
Elevations in transaminases or bilirubin >5x upper limit of normal, or >3.5x the baseline level, are not typical of hepatic TB-IRIS. If present, clinicians should strongly suspect DILI and follow local or national guidelines for hepatotoxicity from anti-tuberculosis and anti-retroviral medications.
The presence of hypersensitivity rash, lactic acidosis, significant coagulopathy or eosinophilia is strongly suggestive of DILI and clinicians should follow local or national guidelines for hepatotoxicity from anti-tuberculosis and anti-retroviral medications.
In cases where anti-tuberculosis and/or anti-retroviral medications have been discontinued but liver enzymes are failing to resolve, liver biopsy should be considered if available.
Features suggestive of hepatic TB-IRIS as a cause of jaundice include:
evidence of IRIS in another system;
new tender hepatomegaly;
predominantly cholestatic liver test derangement;
abdominal ultrasound findings of diffusely abnormal liver texture, lymphadenopathy and normal biliary tree.
If diagnostic uncertainty exists clinicians should consider liver biopsy, if available, or transfer to a higher level of care.
Histopathological findings which suggest TB-IRIS in liver tissue include large epithelioid granulomatous inflammation. Features which are atypical of conventional tuberculous response – including neutrophils, plasma cells and large numbers of lymphocytes – may also be present.
Multiple causes of liver disease are common in HIV-TB co-infected patients presenting with jaundice following HAART initiation, and additional diagnoses such as bacterial sepsis, DILI from traditional medicines, and active hepatitis B infection should be actively sought.
Causes of liver disease in HIV-TB co-infected patients show wide variation in relative frequency according to geographical location and population studied; clinicians should be aware of local and national experience for their setting.