Beginning in May 2003, after receiving written informed consent we assessed adults aged ≥18 with HIV-1 who were clients of the Tororo Branch of the AIDS Support Organization (TASO) for eligibility for study enrolment. Enrolment was offered to clients with a CD4 cell count <250 cells × 10⁶/L or severe HIV disease (defined as WHO stage 3 or 4 or a history of recurrent herpes zoster), a serum aspartate aminotransferase (AST) or alanine aminotransferase (ALT) activity less than five times the upper limit of normal, a calculated creatinine clearance ≥0.42 mL/s, and a Karnofsky score of >40%; isolated pulmonary tuberculosis alone was not an inclusion criterion because in Africa it is common among people with HIV even without immunocompromise. Participants who had taken ART previously were enrolled in the study, but unless use consisted only of nevirapine for prevention of mother to child transmission they were excluded from analyses.

Investigators used a random number generator and blocked randomisation to assign participants to one of three different monitoring regimens: the “viral load arm,” with quarterly HIV viral loads and CD4 cell counts and weekly home visits by a trained lay person using a standardised symptom questionnaire; the “CD4 arm,” with quarterly CD4 cell counts and weekly home visits; or the “clinical arm,” with weekly visits alone. All participants had measurements of viral load and CD4 cell count performed on quarterly blood samples. Results of these were made available to study physicians as appropriate to the participants’ assignment arm. All participants underwent a CD4 cell count at study enrolment to determine eligibility for ART. Follow-up continued until 4 February 2007.

The first line ART regimen was stavudine, lamivudine, and nevirapine (or efavirenz for those taking concurrent rifampicin). Trained lay field officers visited clients’ homes weekly to deliver drugs and collect data regarding drug adherence, potential symptoms of drug toxicity, or death of a household member in the preceding seven days.20 Prepackaged drugs were replaced by using a weekly storage container, and pill counts were conducted at the study clinic by a pharmacist. Participants were weighed each month during home visits and these weights and body mass index (BMI) scores were provided to clinicians. After enrolment, no routine clinic visits were scheduled but participants were encouraged to come to the clinic or hospital if they were ill and were transported to the clinic for assessment if they had specifically defined symptoms or severe illness during a home visit.

Cryptococcal disease was diagnosed by compatible symptoms and serum cryptococcal antigen testing (Crypto-LA, Wampole Laboratories, Cranberry, NJ). Pulmonary tuberculosis was defined as positive results for acid fast bacilli during microscopy from two sputum smears or negative results from sputum smears but with chest radiography compatible with tuberculosis and a lack of response to a two week trial of antibiotics. Extrapulmonary tuberculosis was diagnosed by clinical presentation and occasionally by lymph node biopsy and pathological confirmation. Diagnoses of Kaposi’s sarcoma and cervical cancer were based on biopsy results. Pneumocystis jirovecii pneumonia was diagnosed clinically with chest radiography, clinical presentation, and a response to cotrimoxazole treatment. Monitoring and diagnostic procedures for the occurrence of illness did not differ between study arms. Physicians responsible for patients in the two study arms that included routine viral loads or CD4 counts, or both, received these results on a quarterly basis.

Participants continued with daily cotrimoxazole prophylaxis regardless of CD4 cell count. Participants who met clinical or laboratory definitions of failure of ART were generally switched to didanosine, tenofovir, and lopinavir/ritonavir.

In the viral load arm, treatment failure was defined as two consecutive viral load measurements >500 copies/mL occurring more than six months after the start of ART. For the CD4 monitoring arm, persistently declining CD4 cell counts on two consecutive measurements was considered to indicate treatment failure. For the clinical arm, predefined primary indications for treatment change included determination of a new or recurrent CDC category C condition after three months of ART, documented unintentional weight loss of ≥10%, constitutional symptoms such as fever (>38.5°C) or diarrhoea lasting more than a month without a correctable cause, or new oral or recurrent vaginal candidiasis. Pulmonary tuberculosis itself was not an indication for failure. The first response to a worsening trend in CD4 or viral load was counselling about adherence to treatment. Supplemental CD4 and viral load assessments could be ordered during clinic visits. Study physicians, nurses, counsellors, and other staff met weekly in a case conference and discussed all deaths, opportunistic illnesses, and abnormal laboratory results and approved all regimen changes.

Plasma samples were screened for HIV infection by two enzyme linked immunoassays (EIA) in parallel (Recombigen HIV-1/HIV-2, Trinity Biotech, Dublin; and Murex HIV1.2.0, Abbott diagnostics, Chicago, IL). Specimens concordantly positive or negative received no further testing. Specimens with discordant results were retested with the same algorithm and, if still discordant, by western blotting (LAV Blot, Biorad, Richmond, VA). HIV viral loads were measured with Cobas Amplicor HIV-1 Monitor version 1.5 ultrasensitive assay (Roche, Branchburg, NJ) for baseline measurements, which had a lower limit of detection of 400 copies/mL. Follow-up viral load measurements were conducted with the Cobas Taqman (manual extraction) assay, with a lower limit of detection of 50 copies/mL. CD4 cell counts were done with TriTEST reagents following an in house dual platform protocol and MultiSET and Attractors software with a FACScan or FACSCalibur flow cytometer (Becton-Dickinson, Franklin Lakes, NJ). Complete blood counts were provided with CD4 cell counts.

The main outcome was a composite of severe AIDS associated morbidity and mortality (referred to as severe morbidity and mortality). This included death from any cause, all US Centers for Disease Control category C illnesses except pulmonary tuberculosis, and four additional category B illnesses: more than one episode of herpes zoster or in multiple dermatomes, recurrent severe bacterial infections, peripheral neuropathy not associated with drug toxicity, or oral hairy leucoplakia. We planned to enrol a minimum of 333 participants in each monitoring arm. With 80% power and a two sided test at a median follow-up of three years, this sample size should allow us to detect an absolute difference of 8% between the proportion of participants with a serious morbidity or who died in the clinical arm and the viral load arm if the proportion in the viral load arm was 12%. Data were entered with Epi Info (CDC, Atlanta, GA) and analysed with SAS 9.0 (SAS Institute, Cary, NC).

For the primary analysis, we used an intention to treat approach including all randomised participants. We also performed per protocol analyses that included only participants who received ART for at least 90 days as the different monitoring regimens would not theoretically affect outcomes until the time of the first blood sample. We developed Kaplan-Meier survival curves to compare outcomes and used multivariable Cox regression models to adjust for possible associations with morbidity and mortality, with a backward stepwise approach. Hazard ratios and incidence rate ratios were adjusted for age, sex, and baseline CD4 cell count, viral load, and BMI, unless otherwise specified. A P value <0.05 was considered significant. We assessed the proportionality assumption of time dependent covariates—that is, interaction between monitoring arms and time, and over- and under- dispersion of data. For analyses of the incidence of illnesses, we used a Poisson regression model with a log link function. Person time for people lost to follow-up or transferred to a different provider was censored at the time of the last home visit at which they received ART. We estimated adherence by calculating a weekly drug possession ratio that incorporated an individual’s pill count data and any lapses in delivery that could have occurred because of drug holidays or being away from home.

A data safety monitoring board reviewed data every three months and was asked to reject the null hypothesis of monitoring arm equivalence if the rate of severe morbidity and mortality in any arm exceeded another by three standard errors of the difference (“Haybittle-Peto” rule).21

With an intention to treat approach from the time of ART randomisation, 126 (11%) participants died: 37 in the viral load arm, 40 in the CD4 arm, and 49 in the clinical arm (table 2); 60/126 (48%) deaths occurred in the first three months. Participants in the clinical arm were more likely to die than those in the viral load arm (4.9 v 3.7 per 100 person years; adjusted hazard ratio 1.57, 1.00 to 2.46) and those in the CD4 arm (4.9 v 4.0 per 100 person years; 1.43, 0.92 to 2.21). There was no significant difference in mortality between the CD4 and viral load arms (1.10. 0.69 to 1.75). In the per protocol analysis, there were similar but non-significant trends (table 2).

At baseline, the median viral load was 216 000 copies/mL (interquartile range 72 000-536 000). During follow-up, 61 (5.5%) participants had viral load measurements >500 copies/mL on two consecutive occasions; 16 (4.6%) in the viral load arm, 26 (7.5%) in the CD4 arm, and 19 (5.4%) in the clinical arm (P=0.56). Of 1047 participants in the per protocol analysis, 28 (2.8%) were changed to a regimen containing a protease inhibitor (table 4). Of the 61 participants with virological failure, 13 were switched to second line regimens: seven (44%) of 16 in the viral load arm, four (15%) of 26 participants in the CD4 arm, and two (11%) of 19 in the clinical arm. All participants who were changed to second line treatment in the viral load and CD4 arms had at least two consecutive viral loads >500 copies/mL; only two (12%) of 17 patients in the clinical arm who were switched to second line had two consecutive viral loads >500 copies/mL. Of these 17 patients, all were switched more than a year after they started ART because of AIDS defining events: six had cryptococcal disease, five had Kaposi’s sarcoma; one had extrapulmonary tuberculosis, two had >10% unintentional weight loss, one had cervical cancer; one had cytomegalovirus infection, and one had recurrent pneumonia. Eight (47%) participants changed in the clinical arm had CD4 cell counts <200 cells × 10⁶/L at the time of their illness.

Among participants with virological failure who were not switched to second line therapy the median viral load six months later was greater for those in the clinical arm (7340 copies/mL, interquartile range 2315-10 350) compared with <50 copies/mL (<50-<50) (P=0.002) in the viral load arm and 1340 copies/mL (<50-9540) (P=0.012) in the CD4 arm. Among the 17 people in the clinical arm with virological failure who were not switched to second line treament, however, only two deaths and no opportunistic infections occurred. For participants in the clinical arm, the mean medication possession ratio was worse in the three months after viral load failure compared with the three months before (94.1% v 98.5% of pills taken; mean pairwise difference 4.5%, 0.4% to 8.5%; P=0.032) but it was not significantly different for the CD4 arm (99.1% v 98.6%; −0.5%, −2.3% to 1.3%; P=0.57) or the viral load arm (92.1% v 100%; 7.9%, −4.2% to 20.1%; P=0.17).

Our results are largely consistent with those of the Development of Antiretroviral Therapy for Africa (DART) trial, which was recently completed in two sites in Uganda and one in Zimbabwe,18 although we differ somewhat in our interpretation of these results. Participants in DART were randomised to clinical monitoring alone or clinical monitoring plus six monthly monitoring of CD4 cell count. After five years of follow-up, survival was 87% (95% confidence interval 85% to 88%) in the clinical monitoring arm and 90% (88% to 91%) in the clinical plus CD4 cell count monitoring arm. Neither group received routine viral load testing. While the authors concluded that these differences were quite small and probably did not justify the added expense of CD4 cell count testing, they were unable to show that clinical monitoring was not inferior to CD4 cell count monitoring. Our associated formal cost effectiveness analysis of our study further informs the debate regarding appropriate resource allocation for ART programmes in this region.43

In this randomised trial we examined the efficacy of both routine CD4 cell count and viral load monitoring on clinical outcomes in a resource limited setting. The study does have some limitations. Although participants were randomised at baseline, there were more men in the clinical arm, and at least one study has found that men have worse clinical outcomes during ART than women, although this was mostly because of more advanced immunodeficiency and worse adherence to treatment.44 To account for this potential issue, we adjusted for sex in all models. The virological criteria for switching were conservative in relation to switching criteria in industrialised countries,45 although more liberal than current WHO recommendations,19 and could have led to underestimation of the effectiveness of viral load monitoring in improving clinical outcomes. Also, because we had few participants with virological failure, we had limited statistical power to examine the effectiveness of viral load testing. Lastly, although diagnostic procedures for opportunistic infections were identical across arms, clinicians could not be blinded, and they might have more intensively examined participants in the clinical arm for illness. Some of the illnesses had clear laboratory or pathological diagnoses, such as cryptococcal disease and Kaposi’s sarcoma, and there was increased mortality in the clinical arm that would not be susceptible to ascertainment bias.

The number of people estimated to be acquiring new HIV infections in the world each year exceeds the number currently taking ART.46 As we expand programmes to meet the current and future need for ART, efforts should be made to increase access to CD4 cell counts not only for determining eligibility for ART but also for monitoring people already receiving ART.