Design

We included nine RCTs, with a total of 5426 participants. Two trials were conducted in Thailand (Chariyalertsak 2002; Chetchotisakd 2004), four in the USA (Goldman 2005; McKinsey 1999; Revankar 1998; Schuman 1997), one in Uganda (Parkes‐Ratanshi 2011), and two were multi‐centre trials conducted in Uganda, Zimbabwe, Malawi, and Kenya (Hakim 2017), and Australia, Canada, South Africa, and the UK (Smith 2001).

Participants

Most trials included both adults and adolescents, older than 13 years. One trial included adolescents over 15 years (Parkes‐Ratanshi 2011). One trial also included children older than five years (Hakim 2017).

Six trials did not report on the cryptococcal antigen (CrAg) status of their participants at baseline (Chariyalertsak 2002; Goldman 2005; McKinsey 1999; Revankar 1998; Schuman 1997; Smith 2001). Chetchotisakd 2004 and Parkes‐Ratanshi 2011 reported on the CrAg status of their participants at baseline, but excluded the CrAg‐positive patients. Hakim 2017 reported on the CrAg status of participants at baseline, but did not exclude the CrAg‐positive patients.

Full inclusion and exclusion criteria are presented in the ‘Characteristics of included studies' table.

Interventions

Six trials randomly assigned HIV‐positive participants to the antifungal study drug or placebo (Chariyalertsak 2002; Chetchotisakd 2004; McKinsey 1999; Parkes‐Ratanshi 2011; Schuman 1997; Smith 2001). Two studies randomized participants to continuous administration of antifungal prophylaxis or antifungals, as needed for the treatment of candidiasis (Goldman 2005; Revankar 1998). Hakim 2017 assigned participants randomly to either standard prophylaxis for Pneumocystis jiroveci pneumonia (PJP) with trimethoprim‐sulfamethoxazole or an enhanced prophylaxis package consisting of 12 weeks of fluconazole (100 mg once a day), one dose of albendazole 400 mg, five days of azithromycin (500 mg once a day), 12 weeks of trimethoprim–sulfamethoxazole (trimethoprim 160 mg once a day and sulfamethoxazole 800 mg once a day), isoniazid 300 mg once a day, and pyridoxine 25 mg once a day for 12 weeks.

The choices and doses of antifungal used included itraconazole 200 mg daily (Chariyalertsak 2002; McKinsey 1999; Smith 2001), fluconazole 100 mg daily (Hakim 2017), fluconazole 200 mg daily (Revankar 1998), fluconazole 200 mg three times per week (Goldman 2005; Parkes‐Ratanshi 2011), fluconazole 200 mg weekly (Schuman 1997), and fluconazole 400 mg weekly (Chetchotisakd 2004).

Five included studies did not report if participants received co‐trimoxazole prophylaxis (Chetchotisakd 2004; Goldman 2005; McKinsey 1999; Revankar 1998; Schuman 1997). Seventy‐five percent of participants in the treatment arm and 65% of participants in the placebo arm received co‐trimoxazole prophylaxis in Smith 2001. One study reported that participants were offered co‐trimoxazole according to national guidelines (Parkes‐Ratanshi 2011). All participants in two trials received standard co‐trimoxazole prophylaxis (Chariyalertsak 2002; Hakim 2017).

Participants in the Hakim 2017 and Parkes‐Ratanshi 2011 trials were all anti‐retroviral therapy (ART)‐naïve at the start of follow‐up, and then received current standard ART triple therapy, initiated during the trial. Participant therapies in the Hakim 2017 trial initiated ART at a median of five days, as would be expected under the current standard of care. Participants in the Parkes‐Ratanshi 2011 study initiated ART at a median of 11 weeks. Five trials included participants that were on a mix of a non‐current standard ART regimen and no ART at baseline (Chariyalertsak 2002; Goldman 2005; McKinsey 1999; Schuman 1997; Smith 2001). HIV‐positive participants in Chetchotisakd 2004 were all ART‐naïve at baseline, but they did not report which ART regimen they initiated. One trial did not report the ART status of its participants (Revankar 1998).

Outcome measures

Seven studies reported death as an outcome (Chariyalertsak 2002; Chetchotisakd 2004; Goldman 2005; Hakim 2017; McKinsey 1999; Parkes‐Ratanshi 2011; Smith 2001); we included six of these studies in our analysis. Hakim 2017 reported all‐cause mortality; however, the co‐interventions used in this study, as described in Table 1, could possibly have confounded any effect measured. Therefore, we did not include this study in our meta‐analysis for this outcome. The CD4 cell count thresholds for initiation of antifungal prophylaxis varied from < 100 cells/µL to < 300 cells/µL. Duration of follow‐up varied from 22 weeks to 42 months.

Seven studies reported the incidence of cryptococcal disease (Chariyalertsak 2002; Chetchotisakd 2004; Goldman 2005; Hakim 2017; McKinsey 1999; Parkes‐Ratanshi 2011; Smith 2001). Six studies measured cryptococcal disease occurrence, and used standard prophylaxis, consisting solely of an antifungal or placebo as an adjunct to standard care.

Five studies reported mortality due to cryptococcal disease (Chariyalertsak 2002; Chetchotisakd 2004; Hakim 2017; McKinsey 1999; Parkes‐Ratanshi 2011). In these studies, there was variable reporting of the method of diagnosis of death due to cryptococcal disease. Hakim 2017 measured cryptococcal disease occurrence and used enhanced prophylaxis, which included co‐interventions, as described in Table 1. We did not deem these co‐interventions to be active on mortality due to cryptococcal disease, and so included this study in the pooled estimate.

Only Chariyalertsak 2002 reported adherence to antifungal prophylaxis.

Four studies reported clinically defined Candida resistance in patients enrolled in trials (Chariyalertsak 2002; Goldman 2005; Revankar 1998; Schuman 1997). Chariyalertsak 2002 compared Itraconazole to placebo, while Goldman 2005, Revankar 1998, and Schuman 1997 compared fluconazole to placebo. We identified four studies that reported microbiologically‐defined resistance in Candida species isolated from patients enrolled in trials (Goldman 2005; McKinsey 1999; Revankar 1998; Schuman 1997).

Four studies reported discontinuation of antifungal prophylaxis compared to placebo for any reason, and adverse events (Chariyalertsak 2002; McKinsey 1999; Parkes‐Ratanshi 2011; Smith 2001).

Studies awaiting classification

We were unable to retrieve the full‐text reports of two studies to assess them for inclusion (Smith 1999, Anonymous 1998).

All‐cause mortality

Antifungal prophylaxis had no consistent effect on all‐cause mortality (risk ratio (RR) 1.07, 95% CI 0.80 to 1.43; six trials, 3220 participants; Analysis 1.1). We could not include data for this outcome from the most recent trial, which initiated ART a mean of five days after screening, as there were co‐interventions in the intervention arm that would have confounded the effect on mortality (Hakim 2017).

Cryptococcal disease occurrence

We excluded unconfirmed, suspected cases of cryptococcal disease from our analysis. Hakim 2017 measured cryptococcal disease occurrence, and used enhanced prophylaxis, which included co‐interventions described in Table 1. We did not deem these co‐interventions to be active on cryptococcal disease, so included this study in the pooled estimate.

The seven studies that measured cryptococcal disease identified 91 cases. Most of the studies did not report the source of the cryptococcal infection, simply referring to invasive cryptococcal disease. All 10 cases in Chetchotisakd 2004 were confirmed cases of cryptococcal meningitis; Smith 2001 reported one case of cryptococcal pneumonia and one case of cryptococcal meningitis. Parkes‐Ratanshi 2011 confirmed 11 cases of cryptococcal meningitis, five participants with invasive cryptococcal disease and positive blood cultures, and three participants who became CrAg‐positive after starting prophylaxis. Hakim 2017 reported 32 new cases of cryptococcal infection: 22 cases of cryptococcal meningitis, and one case of cryptococcal fungaemia in the standard prophylaxis arm, and nine cases of cryptococcal meningitis in the enhanced prophylaxis arm.

Meta‐analysis showed a large reduction in the risk of developing cryptococcal disease in those who received antifungal prophylaxis. Participants on antifungal prophylaxis were 71% less likely to develop cryptococcal disease than those receiving placebo or standard care (RR 0.29, 95% CI 0.17 to 0.49; seven trials, 5000 participants; Analysis 1.7). Benefit of antifungal prophylaxis was seen consistently across the included studies, although this was not statistically significant at a 95% level of confidence in four of the studies.

Cryptococcal‐specific mortality

People taking antifungal prophylaxis were less likely to die from cryptococcal disease (RR 0.29, 95% CI 0.11 to 0.72; five trials, 3813 participants; Analysis 1.13).

No clear difference was seen in studies that excluded participants who tested CrAG‐positive, and those on current standard ART regimens (one nucleoside reverse transcriptase inhibitor and two non‐nucleoside reverse transcriptase inhibitors).

Adherence

Chariyalertsak 2002 (129 participants) reported no significant difference in adherence between participants receiving antifungals and placebo. Ninety‐two per cent of those receiving antifungals adhered to the regimen, while 85% of those receiving placebo adhered.

Cryptococcal antifungal drug resistance

We did not identify any studies that reported cryptococcal antifungal resistance.

Infections caused by Candida species resistant to the prophylactic antifungal drug triazole

(a) Clinical resistance

Schuman 1997 compared fluconazole to placebo for the prevention of candidiasis. Two open label trials compared the continuous use of fluconazole prophylaxis for symptomatic treatment of clinical Candida disease (Goldman 2005; Revankar 1998). Clinical resistance was largely defined as participants who developed Candida disease that did not respond to treatment with fluconazole; the exact definition varied between studies, as described in Table 2. We subgrouped the results of this analysis by antifungal therapy.

(b) Microbiological resistance

Three studies monitored resistance by taking surveillance cultures obtained from mucosal swabs, and reporting all strains of Candida resistant to fluconazole (Goldman 2005; Revankar 1998; Schuman 1997). Goldman 2005 and Revankar 1998 reported resistance in oropharyngeal swabs, and Schuman 1997 reported results from vaginal swabs. One study only reported Candida albicans isolates (McKinsey 1999). McKinsey 1999 used itraconazole, and reported both resistance to itraconazole and cross‐resistance to fluconazole, from swabs of any mucosa, from participants with clinical disease. We defined resistance to fluconazole as a minimum inhibitory concentration (MIC) > 16 µg/mL. All studies reported this. Schuman 1997 reported participants with a MIC > 16 µg/mL as ‘dose‐dependent susceptible'. They reported absolute resistance as MIC > 64 µg/mL. For this analysis, we combined participants with these results to form an aggregate number of events with MIC > 16 µg/mL (Table 3). There was marked qualitative heterogeneity between studies that reported on this outcome, as sampling methods, antifungal drug, and Candida species detected differed markedly between McKinsey 1999 and the remaining studies. As a result, we chose not to pool estimates across all three studies.

Treatment discontinuation

Four studies reported the discontinuation of antifungal prophylaxis compared to placebo for any reason (Chariyalertsak 2002; McKinsey 1999; Parkes‐Ratanshi 2011; Smith 2001). The reasons included serious adverse events, hepatotoxicity, pregnancy, use of contraindicated medications (such as rifampicin), and patient decision (Table 4). We found no clear difference between those who discontinued antifungal prophylaxis compared to placebo (RR 1.01, 95% CI 0.91 to 1.13; four trials, 2317 participants; Analysis 1.16).

Adverse events

We excluded Hakim 2017 from the analysis of adverse events, as unpicking the effects of the co‐interventions delivered in this trial was not possible.

(a) Serious adverse events

Four studies reported serious adverse events (Chariyalertsak 2002; Chetchotisakd 2004; McKinsey 1999; Smith 2001). These were measured as the number of patients experiencing at least one serious adverse event. One study reported no adverse events in either group (Chetchotisakd 2004). All studies were conducted before 2004, and as such, the participants were on a mix of older anti‐retroviral drugs, described in Table 1. There was no clear difference in the occurrence of serious adverse events between participants receiving antifungal prophylaxis and those receiving placebo. (RR 1.08, 95% CI 0.83 to 1.41; four trials, 888 participants; Analysis 1.17)

(b) Any adverse event

Four studies reported any adverse events (Chariyalertsak 2002; McKinsey 1999; Parkes‐Ratanshi 2011; Smith 2001).Three out of the four studies were conducted before 2004, and as such, the participants were on a mix of older anti‐retroviral drugs, described in Table 1. Adverse events were measured as the number of patients experiencing at least one adverse event. There was no clear difference in the occurrence of adverse events between participants receiving antifungal prophylaxis and those receiving placebo (RR 1.07, 95% CI 0.88 to 1.30; 4 trials; 2317 participants; Analysis 1.18).

No clear difference was found between groups for any of the most commonly reported adverse events (Analysis 1.19).

• Diarrhoea (RR 1.31, 95% CI 0.32 to 5.29; 2 trials, 424 participants)

• Abdominal pain (RR 0.91, 95% CI 0.56 to 1.46; 2 trials, 1814 participants)

• Nausea (RR 0.97, 95% CI 0.64 to 1.47; 2 trials, 1814 participants)

• Rash (RR 1.03, 95% CI 0.56 to 1.9; 4 trials, 2317 participants)