Survival after diagnosis of AIDS: a prospective observational study of 2625 patientsBMJ 1997; 314 doi: https://doi.org/10.1136/bmj.314.7078.409 (Published 08 February 1997) Cite this as: BMJ 1997;314:409
- Amanda Mocroft, research statisticiana,
- Michael Youle, consultant physician in HIV/AIDSc,
- Julian Morcinek, information consultantc,
- Caroline A Sabin, lecturer in epidemiology and medical statisticsa,
- Brian Gazzard, consultant physician in HIV/AIDSc,
- Andrew N Phillips, reader in epidemiology and biostatisticsa,
- Margaret Johnson, consultant physician in HIV/AIDSb
- a HIV Research Unit Department of Primary Care and Population Sciences Royal Free Hospital School of Medicine London NW3 2PF
- b Department of Thoracic Medicine Royal Free Hospital
- c Kobler Centre St Stephens Clinic Chelsea and Westminster Hospital London SW10 9TH
- Correspondence to: Miss Mocroft
- Accepted 29 November 1996
Objective: To estimate median survival and changes in survival in patients diagnosed as having AIDS.
Design: Prospective observational study.
Setting: Clinics in two large London hospitals.
Subjects: 2625 patients with AIDS seen between 1982 and July 1995.
Main outcome measures: Survival, estimated using lifetable analyses, and factors associated with survival, identified from Cox proportional hazards models.
Results: Median survival (20 months) was longer than previous estimates. The CD4 lymphocyte count at or before initial AIDS defining illness decreased significantly over time from 90x106/l during 1987 or earlier to 40x106/l during 1994 and 1995 (P<0.0001). In the first three months after diagnosis, patients in whom AIDS was diagnosed after 1987 had a much lower risk of death (relative risk 0.44, 95% confidence interval 0.22 to 0.86; P=0.017) than patients diagnosed before 1987. When the diagnosis was based on oesophageal candidiasis or Kaposi's sarcoma, patients had a lower risk of death than when the diagnosis was based on Pneumocystis carinii pneumonia (0.21 (0.07 to 0.59), P=0.0030 and 0.37 (0.16 to 0.83), P=0.016). Three months after AIDS diagnosis, the risk of death was similar in patients whose diagnosis was made after and before 1987 (1.02 (0.79 to 1.31), P=0.91). There were no differences in survival between patients diagnosed during 1988-90, 1991-3, or 1994-5.
Conclusions: In later years, patients were much more likely to survive their initial illness, but long term survival has remained poor. The decrease in CD4 lymphocyte count at AIDS diagnosis indicates that patients are being diagnosed as having AIDS at ever more advanced stages of immunodeficiency.
Estimates of the prognosis of AIDS patients help with allocation of resources and future research
Historically, surveillance data have been used to estimate survival in large patient groups
In an unselected group of 2625 patients with AIDS, median survival (20 months) was longer than previously estimated; the CD4 count at diagnosis decreased significantly over time
After 1987, patients were much more likely to survive an initial AIDS defining illness, but long term prognosis remained poor
There has been little change in prognosis since 1987; this may be due to AIDS being diagnosed at ever more advanced stages of immunodeficiency
Although much has been published about the survival of patients with AIDS, large studies tend to be based on surveillance data from the United States.1 2 3 4 Although data from surveillance studies can estimate survival in large numbers of patients, such studies are limited to basic data collection and follow up. A large British study based on surveillance data that were published in 1993 reported the bias that can occur as a result of reporting delays and underreporting, which are common problems with surveillance data.5 The estimates of median survival from this study may well have changed because of improvements in treatment or the change, in 1993, of the definition of AIDS used in surveillance.6 Rogers et al have more recently published a study based on British surveillance data describing the influence of covariates such as age and calendar year of diagnosis.7 A current estimate of the prognosis of patients with AIDS would help in planning future research and allocation of resources.
A total of 12 565 cases of AIDS were reported between 1982 and March 1996 in England and Wales; 8713 of these patients (69%) are known to have died.8 Our two hospitals have dedicated HIV units that, between them, have seen 2625 patients with AIDS–over a fifth of those in Britain–between 1982 and July 1995, the end date of the present study. Data were collected prospectively on patients throughout this period, and therefore the aim of our study was to provide an up to date estimate of survival in a large, unselected group of patients, and to identify whether survival has improved over time.
Patients and methods
We included all patients from the Chelsea and Westminster Hospital diagnosed as having AIDS between January 1982 and July 1995 and all patients from the Royal Free Hospital (RFH) diagnosed with AIDS between January 1986 and August 1994. AIDS was diagnosed according to the definition in use at the time. For example, patients receiving a diagnosis of pulmonary tuberculosis in 1994 would be classified as AIDS patients, but if the diagnosis had been made in 1991, before the revision of the surveillance definition,6 they would not be classified as AIDS patients. Demographic data, details of all AIDS defining illnesses, treatment, and immunological data are prospectively collected and maintained on a separate database at each site. Prospective data collection began in 1986 at the Chelsea and Westminster and 1990 at the Royal Free Hospital; retrospective data for all patients with HIV who had ever been seen at either clinic was added at this time. Only the factors known at the time of initial AIDS diagnosis were included in this analysis, and no adjustments were made for CD4 lymphocyte counts during follow up or further AIDS defining illnesses.
Estimates of median survival were obtained by using a lifetable analysis.9 Patient survival was measured from the month of diagnosis of initial AIDS defining illness until death. Patients who did not die during the study were censored at the time of their last clinic attendance. Some patients had been diagnosed as having AIDS before their first visit to either hospital. These patients were included in the study, but their survival was left-truncated (survival was calculated from the time of their initial AIDS defining illness, but they were not included in the risk set until the date they were first seen at either hospital). The relative risks of death were obtained by using Cox proportional hazards models. With the exception of the lifetable analysis, all data were analysed with SAS10; all P values are two sided.
To compare survival in different years, patients diagnosed in 1987 or earlier were placed in one group, representing those to whom treatment was not generally available when AIDS was diagnosed; those whose illness was diagnosed after 1987 were placed in a separate group. Patients whose illness was diagnosed after 1987 were split into further groups, but as survival within each group was similar these were combined. Tests of the proportional hazards assumption showed that the effect of year of diagnosis decreased with time. From the lifetable analysis it was clear that the risk of death in the first three months after AIDS diagnosis for patients diagnosed after 1987 was much lower than that of patients diagnosed before 1987. Therefore the relative risk of death in the first three months after AIDS diagnosis was compared with the risk of death after three months. This was modelled by two separate Cox proportional hazards models. The first estimated the relative risk of death in the first three months of follow up; thus patients who were followed up for more than three months were censored at three months. The second model estimated the relative risk of death after three months. In this model patient follow up started three months after a diagnosis of AIDS, so patients who survived for less than three months or who were lost to follow up within three months were not included. There was no evidence that the proportional hazards assumption did not hold in either model (P=0.15 and P=0.78 respectively).
In total, 2625 patients were diagnosed as having AIDS during the study period: 385 (14.7%) from the Royal Free Hospital and 2240 (85.3%) from the Chelsea and Westminster Hospital (table 1). The median duration of follow up after diagnosis of AIDS was 15.4 (90% range 2.3-52.8) months, during which time 1613 patients (61.5%) died. The first diagnosis of AIDS was made at the Chelsea and Westminster Hospital in 1982; the first diagnosis at the Royal Free was in 1986.
The population was, on average, quite young; the median age at AIDS diagnosis was 35.7 (range 2.1-72.2) years. Male patients were significantly older than female patients (median age 35.9 v 30.2 years; P<0.0001, Wilcoxon test). Patients from the homosexual/bisexual exposure category were the oldest patient group (median age 36.2 years) and intravenous drug users were the youngest (median age 31.7 years). In the Royal Free Hospital cohort the proportion of female patients was significantly higher than in Chelsea and Westminster Hospital cohort (45/382 (11.6%) v 81/2240 (3.6%); P<0.0001, 84χ2 test), and the proportion of patients in the homosexual/bisexual exposure category was significantly lower (252/385 (65.5%) v 1954/2240 (87.2%); P<0.0001, χ2 test). Overall, the proportion of female patients with AIDS increased over time, from 1.3% (four patients) before or during 1987 to 7% (31 patients) in 1994 and 1995 (P<0.0001, χ2 test).
The most common single initial AIDS defining illness was Pneumocystis carinii pneumonia (802 cases, 30.6%), followed by Kaposi's sarcoma (490 cases, 18.7%) and oesophageal candidiasis (382 cases, 14.6%). No other single illness was used for diagnosis in over 100 patients; 243 patients (9.3%) were diagnosed with two or more AIDS defining illnesses simultaneously. Kaposi's sarcoma was significantly more likely to be diagnosed in men than women (477/2497 (19.1%) v 12/125 (3.6%); P<0.0001, χ2 test), as was oesophageal candidiasis (370 (14.8%) v 12 (9.6%); P<0.0001, χ2 test). The median CD4 lymphocyte count within three months of the initial AIDS defining illness, available for 1623 (61.8%) patients, was 56x106/l (90% range 6x106/l-416x106/l), and was significantly higher among patients from the Royal Free Hospital (median 62x106/l v 54x106/l at Chelsea and Westminster; p=0.0169; Wilcoxon test). The CD4 lymphocyte count at the initial AIDS defining illness decreased significantly over time, from 90x106/l in patients whose illness was diagnosed during 1987 or earlier to 61x106/l during 1988-90 (P<0.0001, Wilcoxon test).
Median survival overall was 20 months (table 1); only one in 15 patients remained alive five years after diagnosis. Age was strongly related to survival after diagnosis (fig 1). Median survival in patients aged 25 or less at initial diagnosis was 28 months; in patients aged over 55 at diagnosis it was 10 months.
Figure 2 shows the relation between survival and year of diagnosis. In the first three months of follow up, 168 patients (6.4%) died and 127 (4.8%) were lost to follow up. The survival curves for those in whom AIDS was diagnosed before or during 1987 and after 1987 diverge rapidly; 15.9% of patients (54/339) in whom AIDS was diagnosed before or during 1987 had died within three months of diagnosis compared with 5.2% of patients (19/2286) in whom AIDS was diagnosed after this date. The survival curves converge as survival increases, and two years after diagnosis there is little difference in survival between patients in whom AIDS was diagnosed before or during 1987 and after 1987. As the survival curves converge, patients in whom AIDS was diagnosed after 1987 must be at a slightly higher risk of death after three months than those diagnosed before this time. When we broke the period after 1987 into three groups (1988-90, 1991-3, and 1994-5) we found no differences in survival between the groups either during the first three months or afterwards.
To overcome the problem of the converging survival curves, we divided follow up time into two distinct periods and obtained the relative risk of death in each period, as described in the methods. Table 2 shows the results of a multivariate analysis that included all cofactors in the model. We found no significant differences in survival in either time period according to the hospital of diagnosis or sex. Compared with the risk of death for homosexuals or bisexuals, only patients for whom exposure category could not be determined had a significantly greater risk of death. This increased risk was consistent in both time periods and may correspond to a later presentation in this group of patients (relative risk 3.75 during first three months after diagnosis (95% confidence interval 1.20 to 11.75), P=0.024; relative risk 2.60 after first three months (1.41 to 4.79), P=0.0021). The age effect was strong in both periods and CD4 lymphocyte count was strongly related to risk of death. The relative risk of death for a patient with a 50% lower CD4 count at baseline was 1.41 (1.22 to 1.64; P<0.0001) during the first three months and 1.59 (1.50 to 1.68; P<0.0001) after three months.
In the three months immediately after diagnosis of AIDS, patients whose defining illness was Kaposi's sarcoma or oesophageal candidiasis were at a significantly lower risk of death than those whose defining illness was Pneumocystis carinii pneumonia (0.21 (0.07 to 0.59), P=0.003 and 0.37 (0.16 to 0.83), P=0.016). After this time, patients with either of these diagnoses were at a similar risk of death to patients whose diagnosis was based on Pneumocystis carinii pneumonia. When diagnosis was based on a single disease other than oesophageal candidiasis, Kaposi's sarcoma, or Pneumocystis carinii pneumonia, patients had a higher risk of death in both time periods than did patients whose diagnosis was based on Pneumocystis carinii pneumonia (1.43, P=0.11 and 1.30, P>0.005). Risk of death was also higher, but not significantly higher, for patients whose diagnosis was based on two or more diseases.
In the three months after diagnosis of AIDS, patients in whom AIDS was diagnosed after 1987 had a significantly lower risk of death than patients in whom AIDS was diagnosed before this date (0.44; 0.22 to 0.86). As discussed above, when the diagnosis was made after 1987 the patients had a slightly higher risk of death after three months than when the diagnosis was made before 1987 (1.02; 0.79 to 1.31, P=0.91). To further examine the differential effects in the two periods, we considered the interaction between the main covariates and duration of follow up time, modelled as time dependent covariates. This creates a binary variable that takes the value zero for a patient with less than three months' follow up and switches to a value of one after three months' follow up. The interaction with year of diagnosis (after 1987 or before 1987) was highly significant (P<0.0001), supporting the view that patients whose diagnosis was made after 1987 were at a lower risk of death in the three months after an AIDS diagnosis than were patients whose diagnosis was made after this date.
This large cohort of patients with AIDS has helped to confirm cofactors of disease progression. Table 3 summarises published studies of more than 1000 patients with AIDS,1 2 3 4 5 11 12 13 14 15 16 17 ranked by the number of patients studied. Our study is one of the largest studies based on data collected prospectively at a clinic rather than on surveillance data. The estimate of median survival is longer than has previously been suggested and the proportion of patients alive one and two years after diagnosis is considerably higher than that reported in other studies. In addition to the increase in median survival, the proportion of patients who survived for three years (21.8%; 19.6 to 24.0%) was higher than that found by Lundgren et al (16%) in a large observational study of AIDS patients diagnosed between 1979 and 1989 across Europe.12 Changes in the surveillance definition of AIDS6 18 and improvements in antiretroviral treatment and prophylaxis for Pneumocystis carinii pneumonia19 20 21 are both likely to have contributed to the improved survival seen in this patient group.
Patients for whom exposure category was not known had an increased risk of death, even after confounding variables were adjusted for. These patients may form a unique group who were too ill at presentation to be questioned about risk behaviour. Such patients may present with a wide variety of serious medical problems, or may not be well enough to be offered standard treatment with its associated side effects.22 23
Survival and AIDS defining illnesses
In the three months after diagnosis of AIDS, patients whose diagnosis was based on Kaposi's sarcoma and oesophageal candidiasis were at a significantly lower risk of death than those whose diagnosis was based on Pneumocystis carinii pneumonia, which may suggest that these are milder diseases which can initially be treated and are less likely to be terminal when first diagnosed. This is consistent with results from other studies, where patients with these as the initial AIDS defining illness had the longest median survival.12 14 24 25 After three months, patients with other single AIDS defining illnesses had a significantly higher risk of death. Diseases in this category included lymphomas, toxoplasmosis, cytomegalovirus disease, and infection with non-tuberculosis mycobacterium, all of which have a poor prognosis12 14 15 24 and, with the exception of lymphomas, tend to be diagnosed at lower CD4 lymphocyte counts.
Improvements in survival in later years
Many studies have indicated that survival in AIDS patients has improved over time.2 4 5 12 13 25 Our results do not directly show an increase in survival in later years, but they show that in the first three months after diagnosis, patients whose diagnosis was made in later years had a significantly lower risk of death, and this may be due to improvements in treating the initial AIDS defining illness.26 In 1987 and before, patients often died of their initial AIDS defining illness, and a significant proportion of patients would die within three months. Rothenberg et al3 stated in 1987 that almost 12% of patients died within a month of their initial AIDS diagnosis3; this compares with 3.3% in our patients in whom AIDS was diagnosed after 1987. A later study showed that, before 1987, almost one quarter of patients died within three months of their initial AIDS defining illness; during 1987-90 this proportion had dropped to 14%.5 In our study, from three months after diagnosis there was no difference in the risk of death according to year of diagnosis. Lundgren et al showed that although short term survival was improving during the 1980s, the long term prognosis of patients with AIDS was poor.12
Declining CD4 lymphocyte count at diagnosis of AIDS
In this as in other studies12 27 28 29 the average CD4 lymphocyte count at the initial AIDS defining illness has declined over time, suggesting that AIDS is now being diagnosed later and patients are more immunocompromised when AIDS is diagnosed. In addition, the pattern of AIDS defining illnesses has been changing: Kaposi's sarcoma and Pneumocystis carinii pneumonia have become less common as AIDS defining illnesses, and diseases associated with more advanced immunosuppression, such as cytomegalovirus diseases, have become more common.29 30 This has been attributed to the widespread use of antiretroviral therapy and prophylaxis against Pneumocystis carinii pneumonia,28 29 which has been shown to delay the initial AIDS defining condition.19 20 31 If the onset of AIDS has been delayed substantially then survival after a diagnosis of AIDS may be expected to decrease. Survival after diagnosis of aids was longer in our study than previously reported, which may indicate that the time between seroconversion and death is increasing.
Measuring survival after diagnosis of AIDS depends on recognition of the disease, and improvements in survival over recent years have been attributed to an increased awareness and support for patients with AIDS5 and earlier detection of disease.29 An alternative approach to identify possible improvements in survival time is to monitor survival after a given CD4 lymphocyte count–200x106/l, for example–is reached. Issues such as treatment, AIDS defining illnesses, and the role of potential cofactors in improving survival can also be addressed. This has been discussed in part by the recent study by Enger et al31 and will be further investigated in our cohort of patients.
Participants in the collaborative group are listed at the end of the article.
Members of the Royal Free/Chelsea and Westminster Collaborative Group: Royal Free Hospital: Dr M Atkins, Dr S Bhagani, Dr M Bofill, Dr F Bowen, Mr T Drinkwater, Dr J Elford, Dr V Emery, D Farmer, Professor P Griffiths, Professor G Janossy, Dr M Johnson, Dr C Lee, Dr M Lipman, Professor C Loveday, Dr S Madge, Miss A Mocroft, Dr A Olaitan, Dr A Phillips, Dr C Sabin, N Saint, Dr S Jolles, Dr C Stirling, Dr M Tyrer; Chelsea and Westminster Hospital: Dr S Barton, Dr F Boag, Dr B Gazzard, R Halai, Dr D Hawkins, Dr A Lawrence, J Morcinek, Dr M Nelson, Dr M Youle.
Funding: AM was supported by a grant from the Medical Research Council, United Kingdom (No SPG 9219651).
Conflict of interest: None.