Trends in Mortality From Human Immunodeficiency Virus Infection, 1984–2016: An Autopsy-Based Study

This is a retrospective chart review of autopsy reports of adult patients with HIV infection and AIDS performed at Bellevue Hospital from 1984 to 2016. Autopsy records of AIDS patients prior to 1984 could not be extracted. The autopsies described here were performed with next-of-kin permission via a standard autopsy consent, and included patients who died during admission at Bellevue Hospital as well as cases referred to the Bellevue pathology department from surrounding New York City public hospitals and other referring local hospital systems.

After institutional review board review with exemption, all autopsy reports with the terms HIV, AIDS, human immunodeficiency virus, and acquired immunodeficiency syndrome in the final anatomical diagnoses were reviewed by the authors. The patients' medical records, where available, were also reviewed. Paper charts were reviewed for patients prior to 1994 and electronic medical records from 1994 onwards. Data collection included patient age at death, gender, ethnicity, clinical diagnoses known at time of death, and pathologic findings from autopsy. Although autopsy reports listed a unique cause of death in certain reports, following review it was found that in many cases competing equally efficient causes or lack of anatomic evidence for a clinical cause of death was encountered. Therefore, the pathologic anatomic findings rather than the immediate cause of death was reported so that the spectrum of disease could be tallied.

Specific infections were included in diagnoses if laboratory evidence of their presence was documented in the antemortem documentation or found at autopsy. Conditions were deemed AIDS defining if they were among those opportunistic infections and malignancies included in the 1982 definition (Table 1). Disseminated fungal infections other than candidiasis and cryptococcosis were not considered AIDS-defining infections. Only central nervous system lymphoma and KS were included as AIDS-defining malignancies.

Data were analyzed using SPSS Statistics for Macintosh (version 23.0; IBM Corp), R Software,¹⁸  and R Studio.¹⁹  The frequencies of infections and malignancies were analyzed with the χ² test. Trends over time were analyzed in 4-year time groups as well as pre-AZT (pre-1987), pre-HAART (1988–1995), and post-HAART (1997 onwards). Secular trends in the probabilities (prevalence) of various conditions were estimated by logistic regression analysis (P < .05 considered statistically significant) using polynomials to model curvilinearity. For each independent variable modeled, we performed a preliminary analysis, fitting up to sixth-degree polynomials and selecting as the best model the one with the minimum value of the Akaike information criterion. To assess goodness of fit, we visually compared the curves, fitted using individual events, with annual summaries. Unless noted to be nonsignificant, all models presented in the results and figures fit the data statistically significantly better than a constant probability model using the likelihood ratio test.

We retrieved and reviewed a total of 286 autopsies for patients with HIV and/or AIDS that were performed between 1984 and 2016. Among these, 7 fetuses and 27 children younger than 7 years were excluded from the study. The remaining 252 patients were included in the final analysis. Patients autopsied in 1984 were diagnosed as having AIDS based on the surveillance definition. Starting in 1985 the HIV test was available and used to confirm the infection at Bellevue Hospital. Nine patients after 1985 (1989–2010) did not have laboratory confirmation of HIV in the Bellevue medical records system, but a diagnosis of HIV infection from an outside hospital was reported in the autopsy.

The frequency of autopsies during the study period is shown in Table 2, with the trend in Figure 1. The highest number of autopsies (25) was in 1993. There was also a high number of cases (21) in 2008, related to an intervention to provide autopsy education and follow-up to the intensive care unit residents and faculty that year. Pre-HAART (1984–1995), 157 autopsies were performed, with an average of 13 autopsies per year. Post-HAART (1997–2016), 89 autopsies were done, averaging 4.5 autopsies per year. Eighty-six percent of autopsies during the study period were performed on males (Table 2).

Table 2 shows the average patient age for each year during the study period. Data on age were not available for 5 patients in the study. The oldest known patient was 72 years old, and was autopsied in 1995. Mean age at death per year rose from 35 ± 7 years (7 cases) in 1984 to 54 ± 7 years (5 cases) in 2010, after which mean age as well as number of autopsies trended down, with only 1 46-year-old patient autopsied in 2016 (P < .001). Among the 4 autopsies from 2014 to 2016, all were younger patients aged 26 to 46 years, with low CD4⁺ T-cell counts of fewer than 100 cells/mm³. The overall trend of age at death through the pre-AZT, pre-HAART, and post-HAART periods is shown in a polynomial regression model in Figure 2, A.

The first CD4⁺ T-cell count in an HIV autopsy case was recorded in 1992. By regression analysis, mean CD4⁺ T cells increased from 6 cells/mm³ (95% CI, 9–28 cells/mm³) in 1992 to 64 cells/mm³ in 2016 (95% CI, 32–128 cells/mm³; P = .01). Because the distribution of CD4⁺ T-cell counts was skewed, the trend was evaluated with polynomial regression on log₁₀(CD4), which showed a steady increase over the years (Figure 2, B).

Patient race was documented in a total of 213 autopsies until 2005, but was not electronically retrievable or documented in subsequent years. Among 19 autopsies between 1984 and 1987, the most commonly identified race was white, in 9 patients (47%). Subsequently, however, African American and Hispanic races were predominant. Overall, among 252 autopsied patients, 98 (39%) were African American, 75 (30%) were Hispanic, and 39 (16%) did not have race recorded (Figure 3, A and B).

The prevalence of AIDS-defining opportunistic infections decreased during the study period, from 15 among 19 patients (79%) in 1984–1987 to 15 among 37 patients (41%) in 2008–2011 and 2 of 7 patients (29%) in 2012–2016 (Table 3; Figure 4, A). In the last 4-year cohort, no AIDS-defining infections were observed after 2013. The linear trend for these 4-year cohorts over time was statistically significant (P = .03, Mantel-Haenszel χ² for trend = 4.5, df = 1). Overall, 128 autopsies were identified to have AIDS-defining infections. Coinfections from multiple opportunistic pathogens were common. Pneumocystis jirovecii pneumonia (PCP) was the most frequent opportunistic infection, present in 53 autopsies overall, with the highest frequency during 1984–1987 in 10 of 19 patients (53%). Cytomegalovirus (CMV) infection was the second most common infection, found in 47 autopsies during the study period. The probabilities of PCP, CMV, and Cryptococcus infections decreased during the study period, and Mycobacterium avium complex disease infections among autopsies were sporadic and remained fairly constant over the years. These probabilities were assessed by polynomial regression as shown in Figure 5, A through D.

In contrast to the decline in AIDS-defining infections, the frequency of non–AIDS-defining infections increased during the study period, from 7 of 19 patients (37%) in 1984–1987 to 27 of 37 (73%) in 2008–2011 and 4 of 7 (57%) in 2012–2016 (P = .001, df = 1) (Figure 4, B). These infections included bacteremia, endocarditis, pneumonia, Mycobacterium tuberculosis, hepatitis C, and Clostridium difficile, with nearly all patients having coinfections.

We also assessed the overall prevalence of infections as a direct cause of death, including both opportunistic and nonopportunistic infections, and excluding patients with trauma or other predominant factors leading to death. This frequency was very high in the initial study years—100% during 1984–1987. It declined during the next decade to a nadir of 11 of 21 cases (52%) by 1996–1999, and then increased again, as shown with logistic regression analysis in Figure 4, C.

Hepatitis C infections occurred in a total of 15 of the 252 patients. These were distributed sporadically, although notably 3 out of 4 autopsied patients in 2014–2016 died from hepatitis C cirrhosis.

Using the 1982 Centers for Disease Control and Prevention definition for AIDS, KS and central nervous system lymphoma were considered AIDS-defining malignancies. Kaposi sarcoma was first noted in one autopsy in 1984. Its frequency increased over the years and peaked during 1996–1999 with 4 of 17 autopsies (23.5%), after which the frequency declined. The last case of an AIDS-defining malignancy was KS in 2010. The overall trend of AIDS-defining malignancies during the study period was not statistically significant (Figure 6, A).

Non–AIDS-defining malignancies were seen in no patients or one patient in some years, with a mild increase in frequency during the study period that was not statistically significant (Figure 6, B). Non-Hodgkin lymphoma was the most common non–AIDS-defining malignancy. Other examples of malignancies were mesothelioma, non–small cell lung cancer, gallbladder carcinoma, laryngeal carcinoma, and hepatocellular carcinoma. No case of cervical or anal carcinoma was noted in our autopsies.

The first case where coronary atherosclerosis was noted was in 1988, in a 55-year-old male who died of PCP pneumonia. During 1988–1991, coronary and/or aortic atherosclerosis was noted in 11 of 53 patients (21%), and in 2012–2016 it was found in 5 of 7 patients (71%; P < .001). The significant increase in probability of finding atherosclerosis at autopsy, as assessed with polynomial regression, is shown in Figure 6, C. As the stage and extent of plaques were not well defined in all autopsies, only the presence of atherosclerosis has been reported. Because very few cases of heart disease or episodes of cardiac events occurred, these events could not be analyzed.

Other causes of death found in this study included hemorrhagic shock, cardiac arrest of unclear etiology, myocardial infarction, pulmonary embolism, and necrotizing pancreatitis. There was no overall pattern of increasing deaths from cardiovascular disease among the patients.

Autopsies have played a valuable role in the diagnosis and understanding of infectious diseases, particularly those of public health concern. Despite its value as a diagnostic tool, pathology has been underused and autopsy rates have declined during the past few decades, with the emergence of sophisticated diagnostic methods.¹² 

This is the longest longitudinal autopsy study for HIV patients, beginning soon after the onset of the epidemic in 1984 and continuing through the increasing use of antiretroviral therapy in 2016. In 2000, the University of California at San Diego Medical Center reported on autopsy findings in 390 patients with AIDS during the 1982–1998 period.²⁰  In that study, the number of opportunistic infections found in AIDS patients decreased over time, whereas fungal infections were unchanged and bacterial infections increased. Our literature search failed to reveal any other recent autopsy-based reports.

A decline in HIV-related mortality was recognized following the introduction of protease inhibitors in 1995 and HAART in 1996.^1,9  In our study, the highest number of autopsies was seen in 1993, followed by a steady decline. A similar peak in the annual autopsy rate in 1994, followed by a significant decline, has been reported from San Diego.²⁰  Although these autopsy rates show similarities to HIV mortality trends, a likely confounding factor is the overall decline in autopsy rates in the United States during the past few decades.²¹  We also noted a spike in the number of autopsies at our center in 2008. During this year, the autopsy director of the pathology department had monthly conferences with the intensive care unit team to discuss autopsy findings and encouraged autopsies for the purposes of clarity on disease processes, cause-of-death investigation, and education.

A rise in mean age at death at autopsy has been reported²⁰  up to 1998. In our study, the age at death increased monotonically between 1984 and 2010 (Figure 2, A). The oldest patient autopsied, at 72 years of age, was seen in 1995. The largest mean age at death was seen in the year 2010, 54 ± 6 years. However, after 2011 the mean age at death for autopsied patients actually declined to the third or fourth decade of life, and pathologic findings included bacterial infection with hepatitis C cirrhosis. This reflects the fact that despite the overall advances in HIV care, gaps in access to care remained for some younger patients with HIV who were susceptible even in 2016 to liver failure from hepatitis C cirrhosis, sepsis, and death. Based on our experience, we suggest that during this interval autopsies were not pursued as vigorously for older patients with HIV who died with multiple age-related comorbidities.

When examining trends in AIDS-defining conditions, we used the 1982 Centers for Disease Control and Prevention definition for AIDS,¹⁷  which included only CNS lymphoma and KS as AIDS-defining cancers, and a limited list of infections as noted in Table 1. We used this definition because our data start with AIDS patients from 1984, and we avoided changing definitions during the study period.

The decline in AIDS-defining opportunistic infections as seen in our series, including PCP pneumonia, M avium complex, and CMV infections, was noted in epidemiologic data⁶  as early as 1994. In New York, prophylaxis against PCP infection expanded among AIDS patients in 1985. Decline in the incidence of opportunistic infections was also noted after the wider use of HAART¹⁹  after 1996. Our autopsy data showed the prevalence of PCP among autopsied patients declined sharply from 53% in 1984–1987 and then remained low (Figure 5, A).

Similar to our findings, CMV has been reported as a frequent finding at autopsy for HIV patients.²⁰  It has been postulated that CMV may have a direct effect on B-cell response and be a cofactor that augments HIV pathogenicity. The association of CMV and AIDS is an evolving topic, and recently Yen et al²²  reported that CMV was an independent predictor of ischemic stroke in persons living with HIV.

Similar to the previous report²⁰  from 1998, an increase in the frequency of nonopportunistic bacterial and other infections on autopsy in HIV patients over the years was also found in our study, and this trend continued until 2016.

A decline in AIDS-defining cancers and a rise in non–AIDS-defining cancers in HIV patients in the post-HAART era have been clearly shown.^23,24  One cohort from 2001 to 2006 reported up to 70% of malignancies that occurred in HIV patients to be non–AIDS defining.²³  Surprisingly, in our study KS was not a predominant finding even in the earlier autopsies, and the low frequency did not allow a significant trend to emerge (Figure 6, A). We anticipated a rise in non–AIDS-related malignancies during the study period; however, this trend also was not dramatic (Figure 6, B). It is likely that deaths from malignancies in HIV patients in recent years were not consented for autopsy.

Atherosclerosis, hyperlipidemia, and cardiovascular disease have been recognized to have a rising incidence in HIV patients with HAART,²⁵  not only as comorbidities related to older age, but also as possible adverse effects of antiretroviral drugs.^26,27  Our study showed a notable increase in the frequency of coronary and aortic atherosclerosis during the study period, and this trend persisted even as patient age trended down in later years (Figure 6, C), reflecting the rise in this age-related comorbidity.

As this was a hospital autopsy study, it reflects a subset of deaths from HIV during the study period and may not be representative of trends in mortality in the general HIV population, including those deaths referred for forensic autopsies. The clinical diagnoses present in the autopsy report were prepared by medical teams, but supporting laboratory or imaging data to confirm these diagnoses was not available for review and confirmation. CD4⁺ T-cell counts were not available or were not obtained for all patients prior to death. Data on viral loads and whether patients in the study were on antiretroviral therapy or particular antiretroviral regimens such as protease inhibitors were not available. Some autopsy records listed clinical diagnoses from the medical chart and anatomic pathology findings on autopsy together, so these could not be analyzed separately or compared.

Despite the study's limitations, our autopsy data reflect many trends similar to those seen in population-based studies of HIV/AIDS. Autopsies can be a valuable tool in diagnosis of clinical conditions leading to death when studying trends in disease mortality, particularly in those whose immune compromise makes them susceptible to multiple possible causes.

Jonathan Ralston, MD, from the Department of Pathology, NYU School of Medicine, contributed to the collection of autopsy cases up to 2008.