Prognosis of patients with HIV-1 infection starting antiretroviral therapy in sub-saharan Africa: a collaborative analysis of scale-up programmes

Transcription

1 Prognosis of patients with HIV-1 infection starting antiretroviral therapy in sub-saharan Africa: a collaborative analysis of scale-up programmes Margaret May, Andrew Boulle, Sam Phiri, Eugene Messou, Landon Myer, Robin Wood, Olivia Keiser, Jonathan A C Sterne, Francois Dabis, Matthias Egger, for IeDEA Southern Africa and West Africa Summary Background Prognostic models have been developed for patients infected with HIV-1 who start combination antiretroviral therapy (ART) in high-income countries, but not for patients in sub-saharan Africa. We developed two prognostic models to estimate the probability of death in patients starting ART in sub-saharan Africa. Methods We analysed data for adult patients who started ART in four scale-up programmes in Côte d Ivoire, South Africa, and Malawi from 24 to 27. Patients lost to follow-up in the first year were excluded. We used Weibull survival models to construct two prognostic models: one with CD4 cell count, clinical stage, bodyweight, age, and sex (CD4 count model); and one that replaced CD4 cell count with total lymphocyte count and severity of anaemia (total lymphocyte and haemoglobin model), because CD4 cell count is not routinely measured in many African ART programmes. Death from all causes in the first year of ART was the primary outcome. Findings 912 (8 2%) of patients died in the first year of ART. 822 patients were lost to follow-up and not included in the main analysis; patients were analysed. Mortality was strongly associated with high baseline CD4 cell count ( 2 cells per μl vs <2; adjusted hazard ratio 21, 9% CI 17 27), WHO clinical stage (stages III IV vs I II; 3 4, ), bodyweight ( 6 kg vs <4 kg; 23, 18 3), and anaemia status (none vs severe: 27, 2 36). Other independent risk factors for mortality were low total lymphocyte count, advanced age, and male sex. Probability of death at 1 year ranged from 9% (9% CI 6 1 4) to 2 % ( ) with the CD4 model, and from 9% ( 1 4) to 9 6% ( ) with the total lymphocyte and haemoglobin model. Both models accurately predict early mortality in patients starting ART in sub-saharan Africa compared with observed data. Interpretation Prognostic models should be used to counsel patients, plan health services, and predict outcomes for patients with HIV-1 infection in sub-saharan Africa. Funding US National Institute of Allergy And Infectious Diseases, Eunice Kennedy Shriver National Institute of Child Health and Human Development, and National Cancer Institute. Introduction Expected prognosis when combination antiretroviral therapy (ART) is started is of great importance to patients with HIV-1 and their clinicians, and for planning of health-service provision and treatment guidelines. In developed countries, prognosis for patients starting therapy has been modelled in detail. 1 3 A prognostic model developed by a collaboration of prospective studies from Europe and North America showed that risk of death was dependent on CD4 cell count, HIV-1 viral load, clinical stage, history of injecting drug use, and age. 1,2 Provision of ART has been scaled up in sub-saharan Africa since 24. WHO estimated that million patients had started therapy in this region by the end of Mortality is higher in countries with scarce resources than it is in developed ones, especially in the first year of therapy.,6 However, no prognostic models are available for patients in sub-saharan Africa. CD4 cell count and HIV-1 viral load are important prognostic factors for untreated patients and for those receiving ART, but in many clinics in sub-saharan Africa neither CD4 cell counts nor viral load are routinely measured. In these settings, WHO recommends that clinical stage, or clinical stage with total lymphocyte count, should be used to assess eligibility for ART. 7 Studies from high-income settings 8,9 also reported haemoglobin to be a good predictor of mortality in patients starting therapy, as did a small study in Durban, South Africa. 1 We identified risk factors for death in patients starting ART in four large scale-up programmes in sub-saharan Africa and developed two prognostic models: one including CD4 cell count and another in which CD4 cell count was replaced by measurement of total lymphocyte count and haemoglobin. Methods Patients and cohorts We analysed four large scale-up cohorts in sub-saharan Africa that participate in the International epidemiologic Databases to Evaluate AIDS (IeDEA): Gugulethu 11 and Khayelitsha 12 in Cape Town, South Africa; Lighthouse 13 in Lilongwe, Malawi; and Centre de Prise Lancet 21; 376: Published Online July 16, 21 DOI:1.116/S (1) See Comment page 396 Department of Social Medicine, University of Bristol, Bristol, UK (M May PhD, Prof J A C Sterne PhD); Centre for Infectious Disease Epidemiology and Research, School of Public Health and Family Medicine, University of Cape Town, Cape Town, South Africa (A Boulle MD, L Myer PhD); International Center for AIDS Care and Treatment Programs, and Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, USA (L Myer); The Lighthouse Trust at Kamuzu Central Hospital, Lilongwe, Malawi (S Phiri PhD); CEPREF Clinic and Programme PAC-CI, Abidjan, Côte d Ivoire (E Messou MD); The Desmond Tutu HIV Centre, Institute for Infectious Disease and Molecular Medicine, Faculty of Health Sciences, University of Cape Town, Cape Town, South Africa (Prof R Wood); Institute of Social and Preventive Medicine (ISPM), University of Bern, Bern, Switzerland (O Keiser PhD, Prof M Egger MD); and INSERM U897, ISPED, Université Victor Segalen, Bordeaux, France (Prof F Dabis MD) Correspondence to: Prof Matthias Egger, Institute of Social and Preventive Medicine, University of Bern, Finkenhubelweg 11, CH-312 Bern, Switzerland egger@ispm.unibe.ch Vol 376 August 7,

2 en Charge de Recherches et de Formation (CEPREF) 14 in Abidjan, Côte d Ivoire. These programmes were chosen because of their systematic efforts to trace patients lost to follow-up and ascertain deaths. Eligible patients were not previously exposed to ART, aged 16 years and over, knew their date of ART initiation, and had a recorded sex, age, and baseline CD4 cell count within 6 months before treatment. Patients started therapy between January, 24, and March, 27, and had at least 1 year of potential follow-up. In all sites, patients with a CD4 cell count of fewer than 2 cells per μl (irrespective of clinical stage) or WHO stage IV (irrespective of CD4 cell count) were eligible for ART. In Côte d Ivoire and Malawi, but not in South Africa, patients with WHO stage III disease and a CD4 cell count of fewer than 3 cells per μl were also eligible for ART, in line with WHO guidelines. 7 Institutional review boards in all countries approved the analysis of routinely collected programme data at all sites. Follow-up and ascertainment of deaths The primary outcome was death from all causes in the first year of ART. Patients generally attended clinics monthly to obtain drug supplies, although scheduled medical checkups were less frequent, typically every 3 months during the first year of therapy. Most deaths were reported by family members or hospital wards. Patients who did not return to the clinic were traced. In CEPREF, patients who missed two successive visits were traced by social workers who telephoned patients or relatives, and visited patients homes. In Gugulethu, patients were allocated a therapeutic counsellor who lived in the same community, visited patients at home, and provided counselling and adherence support. In Khayelitsha, patients lost to follow-up who could not be contacted by telephone were visited at home by a clinic nurse, or identified through the South African death registry. The Lighthouse clinic used community support to trace patients, and additional efforts were made to trace all patients lost to follow-up from 24, to March, 26, with home visits. CEPREF (Abidjan, Côte d Ivoire) Gugulethu (Cape Town, South Africa) Khayelitsha (Cape Town, South Africa) Lighthouse (Lilongwe, Malawi) Patients 2117 (19%) 1611 (14%) 4397 (39%) 328 (27%) (1%) Age (years) 3 (3 42) 33 (29 39) 33 (28 39) 36 (31 43) 34 (29 41) Sex Female 161 (74%) 199 (68%) 396 (7%) 1798 (9%) 74 (68%) Male 6 (26%) 12 (32%) 131 (3%) 123 (41%) 399 (32%) WHO clinical stage Data reported 299 (99%) 1611 (1%) 4397 (1%) 271 (89%) 1 88 (97%) Advanced 1712 (82%) 121 (78%) 3637 (83%) 299 (96%) 9199 (8%) Less advanced 387 (18%) 36 (22%) 76 (17%) 12 (4%) 169 (1%) CD4 cell count (cells per μl) 129 ( 218) 11 (48 19) 1 (4 161) 127 ( 214) 111 (48 179) Plasma viral load (log copies per ml) Data reported (<1%) 188 (99%) 2732 (62%) 3 (<1%) 4328 (39%) Median 4 8 (4 4 3) 1 (4 6) (4 ) Haemoglobin (mmol/l) Data reported 2111 (99%) 147 (9%) 2879 (6%) 1 (<1%) 6462 (8%) Median 9 ( 2 6 6) 6 8 ( ) 6 8 ( ) 6 ( 6 7 4) Bodyweight (kg) Data reported 2 (97%) 3997 (91%) 2773 (92%) 882 (79%) Median 2 (46 9) 9 (2 67) 3 (47 6) (49 63) Total lymphocyte count (cells per μl) Data reported 212 (99%) 2374 (4%) 4476 (4%) Median 112 ( ) 131 (86 187) 1394 (93 198) Follow-up (years)* Transferred NR 6 (4%) 67 (2%) 412 (14%) 39 (%) Lost to follow-up 244 (12%) 83 (%) 121 (3%) 374 (12%) 822 (7%) Deaths 18 (9%) 124 (8%) 332 (8%) 271 (9%) 912 (8%) 1-year mortality rate (%) Kaplan-Meier estimate (9% CI) 9 9% ( ) 8 2% ( ) 7 8% (7 8 6) 1 % ( ) 8 9% (8 4 9 ) Data are median (IQR) or number (%) unless otherwise stated. CEPREF=Centre de prise en charge de recherches et de formation. NR=Not recorded. =not applicable. *Maximum follow-up per person was 1 year. See methods for definition of loss to follow-up. Kaplan-Meier estimates of mortality relate to patients remaining in care. Table 1: Antiretroviral treatment programmes with number of patients, characteristics at baseline, follow-up time, status at end of study, and Kaplan-Meier estimates of mortality at 1 year Total 4 Vol 376 August 7, 21

3 A Sex B Disease stage (WHO classification) 2 Male Female III or IV I or II Probabiity of death (%) Probabiity of death (%) Probabiity of death (%) Male Female E C Age (years) HIV-1 RNA (log 1 copies per ml) 4 < < WHO I II WHO III IV < D F CD4 count (cells per μl) < Total lymphocyte count (cells per μl) < <4 4 < < Probabiity of death (%) <4 4 < <6 6 G Bodyweight (kg) 2 <4 4 < < Severe Moderate Mild None H Anaemia Severe Moderate Mild None Figure 1: Kaplan-Meier curves of probability of death in patients starting antiretroviral therapy in sub-saharan Africa, 24 7 (A) Sex, (B) baseline disease stage, (C) age, (D) CD4 cell count, (E) viral load, (F) total lymphocyte count, (G) bodyweight, and (H) extent of anaemia. ART=antiretroviral therapy. Vol 376 August 7, 21 41

4 See Online for webappendix Patients (%) Statistical analysis We used a modified intention-to-treat analysis, ignoring treatment changes and interruptions, with Kaplan-Meier curves and survival models developed based on the Weibull distribution. Time was measured from the date of start of ART to the date of death, last follow-up visit, or end of first year. The follow-up time of patients known to have transferred to another facility in the first year was censored at last visit. Patients who were lost to follow-up in the first year were excluded from main analyses but included in a sensitivity analysis, with follow-up time censored at the Person-years of follow-up Deaths Crude hazard ratio (9% CI) Adjusted hazard ratio (9% CI)* Age (years) (27%) (4%) ( ) 83 ( 71 99) (21%) ( ) 1 1 ( ) 7 (7%) ( ) 1 48 ( ) Sex Male 3279 (32%) Female 72 (68%) ( 62 81) 8 ( 74 98) WHO clinical stage, n=1 4 (97%) Less advanced (I II) 122 (1%) Advanced (III IV) 818 (8%) ( ) 3 4 ( ) CD4 cell count (cells per μl) < (1%) (11%) ( 6 88) 72 ( 6 88) (2%) ( 32 46) 4 ( 33 48) (37%) ( 2 29) 27 ( 23 33) (17%) ( 1 2) 21 ( 17 27) Plasma viral load (log copies per ml), n=4187 (41%) <4 411 (1%) < 1684 (4%) ( ) 1 17 ( ) 292 (%) ( ) 1 3 ( ) Anaemia, n=66 (9%) Severe 3 (8%) Moderate 171 (29%) ( 49 73) 61 ( 73) Mild 226 (38%) ( 28 43) 41 ( 33 ) None 137 (2%) ( 1 26) 27 ( 2 36) Bodyweight (kg), n=8139 (79%) <4 132 (13%) < 119 (1%) ( 43 72) 8 ( 47 73) <6 298 (36%) ( 28 44) 39 ( 32 48) (36%) ( 14 2) 23 ( 18 3) Total lymphocyte count (cells per μl), n=418 (4%) <8 738 (17%) (21%) ( 38 8) 74 ( 9 93) (62%) ( 26 1) 69 ( 6 86) Results from Weibull models based on all patients, with missing values imputed. *Adjusted for age, sex, clinical stage, and CD4 group. Severe=haemoglobin less than mmol/l; moderate=haemoglobin <6 2 mmol/l in women and <6 8 mmol/l in men; mild=haemoglobin 6 2 <7 4 mmol/l in women and 6 8 <8 1 mmol/l in men; none=haemoglobin of 7 4 mmol/l or more in women and 8 1 mmol/l or more in men. Table 2: Characteristics and mortality of treatment-naive patients starting combination antiretroviral therapy and remaining in care in four treatment programmes in sub-saharan Africa final visit. Patients were considered lost to follow-up if their last visit was recorded during the first year of ART and they had at least 6 months of potential additional follow-up until the database of their programme was closed and sent for cleaning and analysis. Data for WHO disease stage, bodyweight, haemoglobin, and total lymphocyte count were not available for all patients. We used multiple imputation to account for missing data. 1 The main analysis was based on all patients apart from those lost to follow up from 2 imputed datasets. Webappendix pp 8 9 gives details of the imputation procedure. For sensitivity analyses we included patients who had no missing data in any of the variables required for a particular analysis (complete-case analysis). Analyses were done with Stata (version 1.). Development of the prognostic model was done on the basis of previous methods. 1,2,16 We examined Kaplan-Meier survival curves for each prognostic factor in the prespecified categories of age (16 29, 3 39, 4 49, years), sex, CD4 cell count (<2, 2 49, 99, 1 199, 2 cells per μl), WHO disease stage (III or IV, I or II), bodyweight (<4, 4 <, <6, 6 kg), anaemia (severe [haemoglobin < mmol/l], moderate [ <6 2 mmol/l in women and <6 8 mmol/l in men], mild [6 2 <7 4 mmol/l in women and 6 8 <8 1 mmol/l in men], none [ 7 4 mmol/l in women and 8 1 mmol/l in men]), total lymphocyte count (<8, , 12 cells per μl), and HIV-1 viral load (<4, 4 <, log 1 copies per ml). Weibull proportional hazards models were used to explore crude and adjusted associations with mortality. Models were fitted to the pooled data with no modelling of cohort effect. We used spline smoothing of the baseline hazard that included indicators for the strata with the worse prognosis to model their steep mortality in the first 3 months of treatment. 17 Prognostic factors were assessed for inclusion if the t statistic associated with their coefficient was more than 2 in the multivariable Weibull model. We assessed different combinations of variables to develop a model that incorporated CD4 cell count, and a second model in which CD4 cell count was replaced by other variables. The Akaike information criterion, which penalises complexity, was used to compare the fit of models. We used a system of internal external cross-validation that fits the model on three cohorts and tests predictions in the fourth cohort, rotating around the excluded cohort. 18 We used the D statistic (averaged across imputed datasets) to assess prognostic separation. 19 We calculated the D statistic for the model fitted on the three cohorts and applied to the fourth cohort (D test ). We also calculated the D statistic in the excluded cohort after re-estimating the model coefficients on that cohort (D r ). The difference (D r D test ) is a measure of the degradation in fit when the model is applied to independent data. We favoured models with a low Akaike information criterion score, high D test, and small difference. We assessed the discrimination of the 42 Vol 376 August 7, 21

5 final models using Harrell s concordance (C) statistic, 16 and calculated the R² measure of explained variation for survival models. 19 Calibration was assessed by comparison of Kaplan-Meier curves of observed mortality with curves estimated from the models. Role of the funding source The sponsor of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. MM, AB, OK, and ME had full access to all the data in the study. ME had the final responsibility for the decision to submit for publication. Results The four cohorts had eligible patients with 998 person years of follow-up within 1 year of start of ART. Table 1 shows numbers of patients, years of followup, and outcomes at 1 year in every cohort. The median CD4 cell count at ART initiation was 111 cells per μl (IQR ) for all patients, 117 cells per μl ( 18) for those alive at 1 year, cells per μl (16 124) for those who died, 98 cells per μl (3 186) for those lost to follow-up, and 119 cells per μl ( 21) for those who transferred out. Patients with CD4 cell counts of 2 cells or more per μl were more likely to be in an advanced clinical stage (17 of 1962, 8%) than were those with counts of cells per μl (388 of 493, 7%). Data were missing for 34 (3%) patients for WHO clinical stage, 4691 (42%) for haemoglobin, 2328 (21%) for bodyweight, and 6677 (6%) patients for total lymphocyte count. 822 (7%) patients were lost to follow-up, thus (93%) patients remained in care during the first year. The estimated cumulative percentage of patients in care who died within 1 year of starting ART was 8 9% (9% CI ); and estimated mortality was 8 4% ( ) when patients lost to follow-up were included. Figure 1 shows the Kaplan-Meier survival estimates by prognostic factor. Mortality was highest in the first 3 months of ART, particularly for patients with poor prognosis. Patients who were male, aged 4 years or older, or had advanced disease had high rates of mortality. CD4 cell count, bodyweight, category of anaemia, and total lymphocyte count were all prognostic with a graded relation to mortality. Table 2 shows crude and mutually adjusted hazard ratios (HRs) with 9% CI from Weibull models. The crude HR for viral load greater than log copies per ml was substantially attenuated on adjustment (table 2), and was therefore not considered further for building of the models. Patients years or older had worse survival rates than did those aged 4 49 years (table 2), but few patients were in the oldest age group, and dichotomisation of age at 4 years rather than years resulted in models with improved properties. We noted little evidence of between-cohort heterogeneity in either covariate effects or baseline hazard and therefore fitted a model across all cohorts with the internal external CD4 model Total lymphocyte and haemoglobin model Age (years) < ( ) 1 36 ( ) Sex Male 1 1 Female 68 ( 8 79) 6 ( 73) Clinical stage* Less advanced 1 1 Advanced 2 72 ( ) 2 96 ( ) CD4 count (cells per μl) < ( 62 94) ( 38 7) ( 28 42) 2 29 ( 22 38) Bodyweight (kg) < ( 48 72) 61 ( 47 79) 9 4 ( 33 48) 4 ( 32 1) 6 24 ( 19 3) 2 ( 19 33) Total lymphocyte count (cells per μl) ( 91) 12 3 ( 43 6) Anaemia Severe 1 Moderate 71 ( 6 91) Mild or none 47 ( 37 61) Results from Weibull models adjusted for all variables shown, based on patients remaining in care, with missing values imputed. =variable not included in this model. *Less advanced=who stage I or II; advanced=who stage III or IV. Severe=haemoglobin less than mmol/l; moderate=haemoglobin <6 2 mmol/l in women and <6 8 mmol/l in men; mild=haemoglobin 6 2 <7 4 mmol/l in women and 6 8 <8 1 mmol/l in men; none=haemoglobin of 7 4 mmol/l or more in women and 8 1 mmol/l or more in men. Table 3: Adjusted hazard ratios for death in the CD4 and total lymphocyte and haemoglobin prognostic models cross-validation system. Baseline survival estimates (adjusted for covariates) in the complete-case analysis were consistent with those from analyses of imputed datasets. The first prognostic model included CD4 cell count, age, sex, WHO clinical stage, and bodyweight (CD4 model); and in the second model, CD4 cell count was replaced by total lymphocyte count and haemoglobin (total lymphocyte and haemoglobin model). Table 3 shows the categories of prognostic variables and corresponding HR for the two models. We combined categories of mild and no anaemia because there were few deaths in patients without anaemia. Associations were consistent across models on the basis of all patients and imputed datasets, the subgroup of patients with complete data, and with inclusion or exclusion of patients lost to follow-up (webappendix p 7). The CD4 model had a Vol 376 August 7, 21 43

6 A CD4 model (%) B Total lymphocyte and haemoglobin model 37 (4%) Probability of death (%) (9%) 1324 (13%) 2464 (24%) 926 (9%) 189 (1%) 2683 (26%) 1 (49%) 478 (46%) Figure 2: Cumulative mortality predicted from prognostic models compared with Kaplan-Meier curves of observed mortality (dotted lines) for five prognostic groups of patients starting antiretroviral therapy in sub-saharan Africa (A) Comparison with CD4 cell count model (B) Comparison with total lymphocyte and haemoglobin model. Number of patients (%) is shown. ART=antiretroviral therapy. For the risk calculator see C statistic of 73, D statistic of 1 3, and R² of 28 8%, and the total lymphocyte and haemoglobin model had a C statistic of 721, and D statistic of 1 26, and R² of 27 %. Figure 2 shows survival predicted by the prognostic models with Kaplan-Meier estimates of observed survival, for five prognostic groups defined by risk, such that about 2% of the deaths happened in every group. Both models produced well separated survival curves and a good fit to the Kaplan-Meier estimates. Figure 3 shows survival predicted by the prognostic models with corresponding Kaplan-Meier curves for every cohort for a group of patients at medium risk (female, age <4 years, advanced disease stage, bodyweight 9 kg, and CD4 cell count of cells per μl, or severe anaemia and total lymphocyte count of cells per μl). For these patient groups, the CD4 model slightly over estimated mortality in Gugulethu and the total lymphocyte and haemoglobin model slightly over estimated mortality in CEPREF and Gugulethu and underestimated mortality in Khayelitsha. Webappendix pp 1 2 shows estimates of cumulative mortality at 3, 6, and 12 months after ART start from the CD4 model, for 16 risk groups defined by sex, CD4 cell count, age, clinical stage, and bodyweight. Webappendix pp 3 6 shows estimates from the total lymphocyte and haemoglobin model, for 288 risk groups defined by sex, extent of anaemia, total lymphocyte count, age, disease stage, and bodyweight. A risk calculator based on the prognostic models is available on the collaboration s website. Estimates for cumulative mortality at 1 year from the CD4 model ranged from nearly 1% to almost 3% (table 4). The estimates from the total lymphocyte and haemoglobin model ranged from almost 1% to nearly 6%. Sensitivity analyses in which the prognostic models were fitted with all patients, censoring follow-up time in those lost to follow-up produced lower estimates of cumulative mortality than did the main analysis, especially in the high-risk groups (table 4). Our models predict that mortality is substantially higher for patients in sub-saharan Africa than for those in developed countries. For example, African men younger than 4 years with advanced disease and a CD4 cell count of 99 cells per μl have an estimated risk of death at 1 year of 21 3%, dependent on their bodyweight (webappendix pp 1 2). The estimated mortality in such men in Europe and North America is 2 4%, depending on whether or not the patient has a history of intravenous drug use. 1 Discussion Both our models had good discriminatory power. CD4 cell count is the best prognostic factor in HIV-1 infection, but many ART programmes in sub-saharan Africa do not have the resources to measure it routinely in all patients. CD4 cell counts can be replaced by haemoglobin and total lymphocyte counts for prognostic purposes. We recorded a higher mortality in men than women and therefore produced sex-specific estimates of cumulative mortality. Women were younger and started treatment with less-advanced disease than men, in line with previous analyses, 2,21 which could be attributable to differences in health-seeking behaviour and women s more timely access to ART through antenatal care. Mortality differences persisted with adjustment for age, CD4 cell count, and clinical stage. Good adherence to therapy in women and differences in mortality from other causes, especially violence, could also have an effect. Gender might not be prognostic in other settings, for example where women and men have equal access to ART, and differences might be reversed where men are 44 Vol 376 August 7, 21

7 A CD4 model CEPREF (Côte d Ivoire) 1 Predicted Observed Gugulethu (South Africa) 1 Probability of death (%) Khayelitsha (South Africa) 1 Lighthouse (Malawi) 1 B Total lymphocyte and haemoglobin model 1 CEPREF (Côte d Ivoire) Gugulethu (South Africa) 1 Probability of death (%) Khayelitsha (South Africa) 1 1 Lighthouse (Malawi) Figure 3: Cumulative mortality predicted from prognostic models compared with Kaplan-Meier curves of observed mortality for a group of patients at medium risk of death starting combination antiretroviral treatment in four programmes in sub-saharan Africa (A) Comparison with CD4 cell count model. (B) Comparison with total lymphocyte and haemoglobin model. Patients were female, aged younger than 4 years, had advanced disease stage, weighed 9 kg, and had CD cells per μl, or severe anaemia and a total lymphocyte count of cells per μl. ART=antiretroviral therapy. more likely than women to access therapy (eg, in India or Latin America). 2 Several studies, mainly from North America and Europe, 8,9,22,23 reported that anaemia in patients with HIV-1 infection was associated with high rates of disease progression and death, independent of CD4 cell count and other prognostic factors. Anaemia in HIV-1 infection might be a manifestation of anaemia of chronic disease, infections of the bone marrow, and myelosuppressive drugs. 24,2 Findings from a study in Malawi 26 showed that most nonpregnant adults admitted to hospital with severe anaemia had HIV-1 infection. Our CD4 model would have been improved by addition of anaemia, but applicability would have been reduced because haemoglobin is often not measured routinely in programmes that monitor CD4 cell count. Whereas haemoglobin is prognostic for mortality independent of CD4 cell count, the prognostic value of total lymphocyte count is explained by its correlation with CD4 cell count. Bodyweight was also independent of CD4 cell count as a prognostic measure; bodyweight changes reflect changes in the rate of viral replication, 27 and bodyweight is affected by severe opportunistic infections or malignant diseases. 28 Accuracy of prognostic models can decrease when they are applied to independent data. 29 We kept this effect to a minimum by choosing the model that generalised best to cohorts omitted from the estimation procedure. 18 We used data from four public-sector programmes in three Vol 376 August 7, 21 4

8 CD4 model Total lymphocyte and haemoglobin model Main analysis Analysis including patients lost to follow-up Main analysis Analysis including patients lost to follow-up Highest risk group* 2 % ( ) 4 8% ( ) 9 6% ( ) 3 8% ( ) Lowest risk group 9% ( 6 1 4) 9% ( 6 1 3) 9% ( 1 4) 9% ( 6 1 4) Data are cumulative mortality (9% CI). *Defined as men aged 4 years or older with stage III IV disease, bodyweight less than 4 kg, and CD4 cell counts fewer than 2 cells per μl in the CD4 model; and men aged 4 years or older with stage III IV disease, bodyweight less than 4 kg, total lymphocyte count fewer than 8 cells per μl, and severe anaemia in the total lymphocyte and haemoglobin model. Defined as women aged younger than 4 years with stage I II disease, bodyweight 6 kg or more, and CD4 cell count 2 cells or more per μl in the CD4 model; or women aged younger than 4 years with stage I II disease, bodyweight 6 kg or more, total lymphocyte count 12 cells per μl or more, and mild or no anaemia in the total lymphocyte and haemoglobin model. Table 4: Estimates of cumulative mortality at 1 year in highest and lowest risk groups from main analysis and analysis including patients lost to follow-up. countries, including two cohorts from townships in South Africa. Patients were men and women, and ranged from teenagers to elderly people. Most patients had evidence of clinical disease, but all CD4 cell count categories were well represented. We were interested in prognosis during the scale-up of ART and therefore restricted analyses to patients starting therapy after 24. All programmes provided free therapy a previous analysis showed that free access to treatment was associated with lower mortality than in clinics where patients were charged for treatment. The programmes in this study represent best practice in urban settings, and are not representative of all treatment programmes in sub-saharan Africa. We analysed mortality from all causes, so estimates for patients with stage I II infection might not be applicable to regions that have different rates of mortality unrelated to HIV-1. Nevertheless, results from our models should be applicable to many patients treated in public scale-up clinics in sub-saharan Africa. Our study has several limitations. Data for haemoglobin and total lymphocyte count were absent for many patients. We imputed data that were missing so we could include all patients, avoid selection bias, and increase applicability. 1,3 Associations with prognostic factors from the complete-case analyses were consistent with those from the analyses with imputed datasets. Nevertheless, the improved discrimination of the CD4 model might be attributable to inclusion of CD4 cell count, or because more data were imputed in the total lymphocyte and haemoglobin model than the CD4 model. Data for symptomatic disease, tuberculosis, and other opportunistic infections were not available, but probably affect prognosis. 11 We used bodyweight in construction of the models, instead of body-mass index, because many clinics did not measure height. Furthermore, bodyweight measurements were taken from patients only at the start of ART, but weight lost before initiation and gained after affect survival. 28,31 Patients with a CD4 cell count of more than 2 cells per μl seemed to have an equivalent prognosis to those with counts of 1 2 cells per μl, which is probably explained by confounding by indication. WHO recommends initiation of ART for patients with CD4 cell counts of fewer than 2 cells per μl (except for patients with advanced disease, in whom treatment is recommended at <3 cells per μl, or for those with very advanced disease, in whom treatment is recommended irrespective of count). 7 People with CD4 cell counts of more than 2 cells per μl were more likely to be in an advanced clinical stage than were those with counts of cells per μl. We excluded patients lost to follow-up to reduce bias caused by underascertainment of deaths. Mortality in patients lost to follow-up who were then traced was 4 % in public ART programmes in sub-saharan Africa, 32 which is much higher than was mortality of patients remaining in care (9% at 1 year in this analysis). Previous analyses generally censored follow-up time at the last visit for those lost to follow-up. Compared with the main analysis estimated mortality was reduced in a sensitivity analysis that included patients lost to followup because additional follow-up time, but not deaths, was added to the data. Differences in estimated mortality in the sensitivity analysis, compared with the model excluding patients lost to follow-up, were modest overall but greatest in the groups at highest risk of death. However, exclusion of patients lost to follow-up did not completely remove bias, because patients might have died in care before they met the definition for loss to follow-up. Furthermore, mortality might have been higher in patients whose follow-up was censored (but who did not meet criteria for loss to follow-up) than it was in otherwise identical patients whose follow-up time was not censored (termed informative censoring). The Development of AntiRetroviral Therapy in Africa (DART) study 33 a randomised trial comparing clinical monitoring with clinical and laboratory monitoring (haematology, biochemistry, and CD4 cell counts) concluded that CD4 cell counts should be routinely monitored from the second year of therapy to guide the switch to second-line treatment. CD4 cell counts might therefore not be needed at start of ART, or in the first year of therapy. Furthermore, haemoglobin and total lymphocyte count could replace CD4 cell count for monitoring of patients in the second year of therapy, and for patients not yet eligible for therapy. 34 The role of different laboratory examinations in the scale-up of ART in resource-scarce settings should be clarified, including point-of-care tests for viral load that are in development. Although our study does not establish the CD4 cell count at which therapy should be started, the prognosis of 46 Vol 376 August 7, 21

9 many patients would be improved with more timely start of ART. Expansion of public health strategies to allow early access to therapy in sub-saharan Africa is urgently needed. Contributors MM and ME designed and coordinated the analyses. MM did statistical analyses and developed and validated prognostic models. JACS and OK advised on statistical analyses. ME and MM wrote the first draft of the paper. AB, SP, EM, LM, and RW assisted in implementation, fieldwork, and data collection at study sites. All authors contributed to the final text. ME and AB (Southern Africa) and FD (West Africa) are the principal investigators of the IeDEA regions in the study. Conflicts of interest We declare that we have no conflicts of interest. Acknowledgments We thank all patients, doctors, and study nurses associated with the four cohort studies. The International epidemiologic Databases to Evaluate AIDS (IeDEA) is supported by the National Institute of Allergy And Infectious Diseases (NIAID) (grant 1U1AI69924 to the Southern African Region and U1AI69919 to the West African Region), the National Cancer Institute (NCI) and the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD). MM and JACS were supported by the United Kingdom Medical Research Council (grant G782), FD was supported by the French Agence Nationale de Recherches sur le SIDA et les Hépatites Virales (grants ANRS 1211 and 12138) and ME was supported by the Swiss National Science Foundation (grant 32473B ) and the Office of AIDS Research of the National Institutes of Health. References 1 Egger M, May M, Chêne G, et al. Prognosis of HIV-1-infected patients starting highly active antiretroviral therapy: a collaborative analysis of prospective studies. Lancet 22; 36: May M, Sterne JA, Sabin C, et al. Prognosis of HIV-1-infected patients up to years after initiation of HAART: collaborative analysis of prospective studies. AIDS 27; 21: Lundgren JD, Mocroft A, Gatell JM, et al. A clinically prognostic scoring system for patients receiving highly active antiretroviral therapy: results from the EuroSIDA study. J Infect Dis 22; 18: WHO. Towards universal access. Scaling up priority HIV/AIDS interventions in the health sector, 29 Progress Report. (accessed June 23, 21). The Antiretroviral Therapy in Lower Income Countries (ART-LINC) Collaboration and ART Cohort Collaboration (ART-CC) groups. Mortality of HIV-1-infected patients in the first year of antiretroviral therapy: comparison between low-income and high-income countries. Lancet 26; 367: Keiser O, Orrell C, Egger M, et al. Public-health and individual approaches to antiretroviral therapy: township South Africa and Switzerland compared. PLoS Med 28; : e WHO. Scaling up antiretroviral therapy in resource-limited settings: treatment guidelines for a public health approach, 23 revision. Geneva: World Health Organization, Mocroft A, Kirk O, Barton SE, et al, for the EuroSIDA study group. Anaemia is an independent predictive marker for clinical prognosis in HIV-infected patients from across Europe. AIDS 1999; 13: Harris RJ, Sterne JA, Abgrall S, et al. Prognostic importance of anaemia in HIV type-1-infected patients starting antiretroviral therapy: collaborative analysis of prospective cohort studies. Antivir Ther 28; 13: Ojikutu BO, Zheng H, Walensky RP, et al. Predictors of mortality in patients initiating antiretroviral therapy in Durban, South Africa. S Afr Med J 28; 98: Lawn SD, Harries AD, Anglaret X, Myer L, Wood R. Early mortality among adults accessing antiretroviral treatment programmes in sub-saharan Africa. AIDS 28; 22: Boulle A, Van Cutsem G, Hilderbrand K, et al. Seven-year experience of a primary care antiretroviral treatment programme in Khayelitsha, South Africa. AIDS 21; 24: Beadles WI, Jahn A, Weigel R, Clutterbuck D. Peripheral neuropathy in HIV-positive patients at an antiretroviral clinic in Lilongwe, Malawi. Trop Doct 29; 39: Toure S, Kouadio B, Seyler C, et al. Rapid scaling-up of antiretroviral therapy in 1, adults in Côte d Ivoire: 2-year outcomes and determinants. AIDS 28; 22: Clark TG, Altman DG. Developing a prognostic model in the presence of missing data: an ovarian cancer case study. J Clin Epidemiol 23; 6: May M, Royston P, Egger M, Justice AC, Sterne JA. Development and validation of a prognostic model for survival time data: application to prognosis of HIV positive patients treated with antiretroviral therapy. Stat Med 24; 23: Royston P, Parmar MK. Flexible parametric proportional-hazards and proportional-odds models for censored survival data, with application to prognostic modelling and estimation of treatment effects. Stat Med 22; 21: Royston P, Parmar MK, Sylvester R. Construction and validation of a prognostic model across several studies, with an application in superficial bladder cancer. Stat Med 24; 3: Royston P, Sauerbrei W. A new measure of prognostic separation in survival data. Stat Med 24; 23: Braitstein P, Boulle A, Nash D, et al. Gender and the use of antiretroviral treatment in resource-constrained settings: findings from a multicenter collaboration. J Womens Health (Larchmt) 28; 17: Cornell M, Myer L, Kaplan R, Bekker LG, Wood R. The impact of gender and income on survival and retention in a South African antiretroviral therapy programme. Trop Med Int Health 29; 14: Moore RD, Keruly JC, Chaisson RE. Anemia and survival in HIV infection. J Acquir Immune Defic Syndr Hum Retrovirol 1998; 19: Semba RD, Shah N, Klein RS, Mayer KH, Schuman P, Vlahov D. Prevalence and cumulative incidence of and risk factors for anemia in a multicenter cohort study of human immunodeficiency virusinfected and -uninfected women. Clin Infect Dis 22; 34: Moses A, Nelson J, Bagby GC Jr. The influence of human immunodeficiency virus-1 on hematopoiesis. Blood 1998; 91: Volberding PA, Levine AM, Dieterich D, Mildvan D, Mitsuyasu R, Saag M. Anemia in HIV infection: clinical impact and evidencebased management strategies. Clin Infect Dis 24; 38: Lewis DK, Whitty CJ, Walsh AL, et al. Treatable factors associated with severe anaemia in adults admitted to medical wards in Blantyre, Malawi, an area of high HIV seroprevalence. Trans R Soc Trop Med Hyg 2; 99: Mwamburi DM, Wilson IB, Jacobson DL, et al. Understanding the role of HIV load in determining weight change in the era of highly active antiretroviral therapy. Clin Infect Dis 2; 4: Mangili A, Murman DH, Zampini AM, Wanke CA. Nutrition and HIV infection: review of weight loss and wasting in the era of highly active antiretroviral therapy from the nutrition for healthy living cohort. Clin Infect Dis 26; 42: Justice AC, Covinsky KE, Berlin JA. Assessing the generalizability of prognostic information. Ann Intern Med 1999; 13: Sterne JA, White IR, Carlin JB, et al. Multiple imputation for missing data in epidemiological and clinical research: potential and pitfalls. BMJ 29; 338: b Madec Y, Szumilin E, Genevier C, et al. Weight gain at 3 months of antiretroviral therapy is strongly associated with survival: evidence from two developing countries. AIDS 29; 27: Brinkhof MW, Pujades-Rodriguez M, Egger M. Mortality of patients lost to follow-up in antiretroviral treatment programmes in resource-limited settings: systematic review and meta-analysis. PLoS One 29; 4: e Mugyenyi P, Walker AS, Hakim J, et al. Routine versus clinically driven laboratory monitoring of HIV antiretroviral therapy in Africa (DART): a randomised non-inferiority trial. Lancet 21; 37: Lau B, Gange SJ, Phair JP, Riddler SA, Detels R, Margolick JB. Use of total lymphocyte count and hemoglobin concentration for monitoring progression of HIV infection. J Acquir Immune Defic Syndr 2; 39: Vol 376 August 7, 21 47