II. Methods

Adult and Adolescent Spectrum of HIV Disease Surveillance Project. We analyzed data from January 1990 (when the project began) through August 1996 from the records of patients in the Adult and Adolescent Spectrum of HIV Disease Surveillance Project, a multicenter medical record review in nine United States cities.⁽²⁴⁾ Each project site developed procedures to ensure patient confidentiality and conducted the project according to guidelines of local human subjects review committees. Depending on the project site, either all patients infected with HIV who attended participating clinics or a systematic sample of patients who attended were eligible for observation, and HIV-infected patients with and without AIDS were observed. A standardized instrument was used to collect information on the patient's clinical conditions, laboratory values, and treatments during the year preceding the first medical record abstraction (baseline review); similar information was collected by medical record review every 6 months until the patient's death or last contact (semester abstractions). For each semester abstraction, we used the most recent hemoglobin determination in the semester.

Definitions. Prevalent hemoglobin concentration was defined as the first recorded hemoglobin concentration recorded after first observation in Adult and Adolescent Spectrum of HIV Disease Project (ASD). Univariate statistics were calculated on prevalent hemoglobin concentrations stratified by sex. Strata of hemoglobin concentrations were created based on sex-specific reference ranges.⁽²⁵⁾ For men strata were <10, 10 to 14, and > or = 14 g/dL; for women strata were <10, 10 to 12, and > or = 12 g/dL. Stratified hemoglobin concentrations are presented by stage of HIV disease.

For all other analyses, anemia was defined as a hemoglobin concentration of less than 10 g/dL or an International Classification of Diseases (ICD-9) diagnosis code of 280 to 281.99 (iron deficiency anemia and other deficiency anemias), 283 to 284.79 (acquired hemolytic anemias), 648.2 to 648.29 (anemia complicating pregnancy or childbirth), or 284.9 to 285.99 (aplastic anemia, other than drug related, and other unspecified anemias). Drug-related anemia was defined as a diagnosis of anemia for which the physician specified a drug-related cause in the medical record or for which an ICD-9 code of 284.8 (aplastic anemia caused by drugs) was recorded; all other diagnoses were considered unrelated to drugs.

To determine whether a person with anemia recovered from anemia, hemoglobin levels measured during two semesters after the semester of anemia diagnosis (or one semester, if there was only one semester of follow-up) were compared with the level from the semester of anemia diagnosis. If both hemoglobin levels determined after anemia diagnosis were more than 10.0 g/dL and were at least 1.0 g/dL more than the level at the time of anemia diagnosis, the person was considered to have recovered.

Incidence of anemia. To calculate the incidence of anemia, all anemia diagnoses in the 1-year period from 6 months to 18 months after baseline review (semesters 2 and 3, defined as the study year) were counted; the analysis included all patients who were observed through the end of the study year, died during the study year, or had anemia during the study year. Patients who had only one semester of follow-up, prevalent anemia (defined as anemia during the semester before the study year), or no CD4 count or hemoglobin level during the study year were excluded (Table 1). Multiple logistic regression was performed to describe the associations of demographic variables (sex, age < or > or = 45 years, race, and HIV exposure mode), stage of disease (CD4 T-lymphocyte count of < or > or = 200 cells/µL, and diagnosis of any AIDS-defining opportunistic illness), illnesses (bacterial septicemia, lymphoma, and Mycobacterium avium complex), laboratory data (neutropenia and thrombocytopenia), and chemotherapeutic agents (zidovudine, didanosine, dideoxycytidine, trimethoprim-sulfamethoxazole, ganciclovir, and fluconazole) with the occurrence of anemia during the study year. Drug prescriptions during the semester preceding the semester of anemia incidence, and clinical conditions present during the semester of anemia incidence were examined. Because of the large numbers of observations available, all covariates were kept in the regression model regardless of statistical significance. Separate regression models were constructed for the outcomes of anemia related to drugs and anemia unrelated to drugs. For persons who did not develop anemia, semester 2 was considered the referent semester to determine covariates. The results are reported as adjusted odds ratios (AORs), with 99% confidence intervals (CIs). We used 99% CIs because we performed a large number of tests of significance.

Mortality. Analyses of mortality were conducted from the first CD4 T-lymphocyte count after baseline review ("first CD4 count") to the date of death, last contact, or loss to follow-up. For 75% of patients, the first CD4 count was done in the first semester after baseline review; for 90%, the first CD4 count was done in the first two semesters after baseline review. Patients were excluded if they (1) did not have a blood hemoglobin level and CD4 count at any time after baseline review, (2) developed anemia before the first CD4 count, or (3) had no follow-up after the date of the first CD4 count (Table 2).

The Kaplan-Meier method was used to describe the median survival of patients after the first CD4 count, and the log rank test was used to determine differences in survival between those persons who had (l) no anemia, (2) drug-related anemia, or (3) anemia unrelated to drugs. Proportional hazards regression was conducted to evaluate the effect of anemia on survival from the first CD4 count. The time-dependent regressors were clinical AIDS, antiretroviral therapy (at least one therapy of zidovudine [ZDV], didanosine, or dideoxycytidine), Pneumocystis carinii pneumonia prophylaxis (at least one therapy of aerosolized pentamidine, trimethoprim-sulfamethoxazole [TMP-SMX], or dapsone), neutropenia (white blood cell count of <2,500/µL), thrombocytopenia (platelet count of <50,000/µL or physician diagnosis), and anemia. Non-time-dependent regressors were sex, age (<25, 25 to 44, or > or = 45 years), and HIV exposure mode. Separate analyses were conducted for subgroups stratified by first CD4 counts. The results of these analyses are reported as risk ratios with 99% CIs.

Effect of Recovery from anemia on mortality. In analyses to describe the effects of recovery from anemia on mortality, patients were excluded if they (1) did not have a diagnosis of anemia after first CD4 count, (2) did not have at least one measurement of hemoglobin level in a semester after the incident semester, or (3) had no follow-up after the date of the first CD4 count. The Kaplan-Meier method and proportional hazards regression were conducted as already described to determine the effect of recovery from anemia on mortality after the first CD4 count. Separate analyses were conducted for subgroups stratified by first CD4 counts. The results of these analyses are reported as median survival of patients after first CD4 count and as risk ratios with 99% CIs.