XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
Int J STD AIDS OnlineFirst, published on February 17, 2015 as doi:10.1177/0956462415571970
Original research article
Cause of death in HIV-infected patients in South Carolina (2005–2013)
International Journal of STD & AIDS 0(0) 1–8 ! The Author(s) 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0956462415571970 std.sagepub.com
Michael Cima1, R. David Parker1, Yasir Ahmed2, Sean Cook3, Shana Dykema4, Kristina Dukes3, Stephan Albrecht3 and Sharon Weissman3
Abstract The life span of persons with HIV has been greatly extended over the past 30 years due to novel therapies. In the developed world and urban settings, this results in a lifespan rivaling the lifespan of a person without HIV. A retrospective study was conducted on 459 patients of an urban, academic medical center who died between 2005 and 2013 in a medium-sized US city. Using the established Cause of Death Project (CoDe) protocol, we measured multiple factors including comorbidities, risk behaviours, contributing and underlying causes of death. This study is one of the few USbased studies using this validated protocol. Among the deaths, 25.9% were sudden and 15.2% were unexpected. Almost one-fifth were related to AIDS-related infections; 47.5% related to non-AIDS causes; with the remainder unknown. Statistically significant increases in CD4 counts and decreasing viral loads were observed over the study period. There were no statistically significant differences observed by HIV risk behaviour, race, gender, age at death, or on antiretorivirals at death. In support of the existing literature, improved HIV management appears to reduce the AIDS-related attributable death among patients observed in this study.
Keywords HIV, AIDS, comorbidity, mortality, cause of death Date received: 13 October 2014; accepted: 18 January 2015
Introduction With the widespread use of antiretroviral therapy (ART), HIV-associated mortality has dramatically declined in the USA and other developed nations.1,2 In the USA, life expectancy for newly diagnosed HIV positive patients increased from 10.5 years to 22.5 years between the years 1996 and 2005.3 A report from the North American AIDS Cohort Collaboration on Research and Design (NA_ACCORD) found that HIV-positive individuals treated in their twenties have a life expectancy into the early 1970’s.4 Numerous other studies support the finding that early initiation of HIV therapy correlates longer life expectancy in HIV positive individuals, approaching life expectancy of other chronic, non-HIV illnesses.5–8 As life expectancy of patients infected with HIV increases, comorbid (non-HIV/AIDS) conditions are more instrumental in the morbidity and mortality of HIV-positive patients who are in clinical care.9–12
Complications of aging are becoming more prevalent. HIV patients are at greater risk for aging complications such as osteoporosis, cancer, smoking-related illness and cardiovascular disease.13,14 In addition, ART therapy may accelerate or exacerbate some of these complications.15,16
1
School of Public Health, West Virginia University, Morgantown, WV, USA 2 Department of Internal Medicine, Texas Tech University, Odessa, TX, USA 3 Department of Medicine, University of South Carolina, Columbia, SC, USA 4 South Carolina Hospital Association, Columbia, SC, USA Corresponding author: Sharon Weissman, University of South Carolina, Department of Medicine, 2 Medical Park, Suite 205, Columbia, SC 29203, USA. Email: [email protected]
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
2
International Journal of STD & AIDS 0(0)
Several studies have shown that an increasing proportion of deaths in HIV-infected patients are attributable to causes not traditionally considered to be related to HIV.9–12,17,18 In several of these studies, the nonHIV/AIDS conditions account for greater than 50% of the deaths.9,10,12 These findings are supported by large cohort studies showing these non-HIV/AIDS contribute significantly to mortality among persons with HIV.17,18 This shift from AIDS-related to non-HIV/ AIDS causes of death (CoD) is most prominent in patients receiving ART and with higher current and initial (nadir) CD4 counts.9,10,12,17,18 Thus a population, who is not receiving the maximum benefit from ART therapy, either because of late stage HIV diagnosis or limited access to care, may not experience this same shift in CoD. Therefore, persons in areas of the USA who are more traditionally impacted by barriers in accessing care and corollaries, such as poverty, homelessness, lack of insurance, substance misuse and mental health issues are most likely at risk, such as persons in the Southeastern states.19,20 Multiple CDC MMWR reports have shown that a large portion of HIV-infected patients in South Carolina (SC) are diagnosed late in their course of HIV infection.21,22 From 2001 to 2005, 41% of newly diagnosed HIV patients in SC were late testers, presenting with an AIDS diagnosis or developed AIDS within one year of their HIV diagnosis.23 Furthermore, in SC a significant proportion of people living with HIV and AIDS live in rural areas where access to care and HIV testing is often a barrier.24–27 Thus, SC might not see the same shift to non-HIV/AIDS CoD. There are no published data on CoD for HIV-infected patients neither in the southern states nor in SC. This information is important for targeting future prevention, treatment efforts and research efforts. The aim of this study is to determine CoD in HIVinfected individuals in SC. The secondary aims are to explore the relationships between various factors, such as patient characteristics, immunologic and viral logic parameters, and high risk behaviours such as drug misuse and cause of death. Lastly, to determine if there are any trends over time in CoD or patient characteristics.
This hospital and clinic provide approximately 50% of all HIV services in the state and 90% of HIV care in an eight county area. Patient care and billing databases were queried to generate a patient list for outpatient care. Inpatient data were collected using the electronic health record. To increase case finding, especially among patients lost to care, a list of patients who were last seen on or before 1 January 2008 was compared to state and national death registries. The Cause of Death Project (CoDe) protocol was employed for the purposes of this study.28 The CoDe project is a multinational collaborative effort to develop a standardised approach to data collection and determination of CoD in HIV-infected patients. The protocol of the original study is publicly available and can be used free of charge. Included in the protocol is a standardised Case Report Form (CRF) and algorithm for determining the cause of death. Patient medical records were reviewed and data on HIV treatment, complications, CD4 counts, HIV viral load, comorbid conditions and cause of death were extracted using the CRF. Additional data, not found on the CoDe CRF, were extracted on HIV risk factors, race, substance use, chronic kidney disease (CKD), malignancy, obesity and ART use through a supplemental data collection form.
Classification of cause of death In accordance with CoDe protocol, two clinicians reviewed various data sources to determine a specific cause of death. If a consensus was reached, then CoD was established. If there was disagreement between these two reviewers, or both coded the CoD as unknown or unclassifiable, the specific case was referred to one or more additional reviewers to arbitrate. If a consensus was still not reached, the case was classified according to majority decision or deemed unknown. As stated in the protocol, autopsy results were used to determine CoD when available. In situations where autopsy results were not available, the cause of death was determined based on review of medical records, interview with patients’ primary medical provider, death certificate, obituary, registry, or family report of cause. Family members were not contacted by study personnel.
Methods Sample and data collection
Data analysis
A retrospective medical record review collected data on patients who died between 1 January 2005 and 31 December 2013. CoD data on HIV-infected patients who received inpatient care and/or outpatient care from an infectious diseases clinic and associated hospital in a medium-sized, Southern city were collected.
To determine unadjusted associations between variables, univariate analyses including Fisher’s Exact Test for categorical data and t-tests or the nonparametric equivalent were used. Non-parametrics were used for more exact tests in instances when the cell sizes were too small to allow standard parametric
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
Cima et al.
3
analyses or in situations where there was a high degree of data variance. Different multivariable regressions determined the presence and statistical strength of relationships in models designed to best fit the data. For analyses which included primarily categorical data with a dichotomous outcome, multivariable logistic regressions were used. Multivariable models allowed analyses to consider the presence and effect of multiple variables within the model. Variables included in these models were: smoking tobacco, excessive alcohol consumption, non-injection illicit drug use (non-IDU), illicit injection drug use (IDU), race, gender, last CD4 count, CD4 nadir, viral load prior to death, age at death and ART at death. Stepwise model reduction used a statistically significant criterion of (p ¼ 0.05) for individual variable removal from the model. However, also included were considerations in the variance, Bayesian Information Criteria (BIC) and effect sizes in conjunction with the law of parsimony determining the best model. Additionally, given the potential for differentiation between clinical and statistical significance in any biomedical model, the investigators ensured that the interactions between variables were measured (as indicated) as well as confounding, effect modification and collinearity. In the case of differences between statistically significant and clinically significant models, a determination was made at that time to determine the best model for fit. R square and adjusted r-square values helped to inform these decisions, as well as the confidence intervals, effect sizes and the clinical knowledge of the investigators as well as the aforementioned methods for goodness of fit of each model to the data. Fractional polynomials were used in conjunction with the appropriate regression were used for continuous variables to determine trends across study years.29,30
Results The characteristics of the sample are presented in Tables 1 and 2. Our total sample size (n ¼ 459) had a median age of 48 years at death with a median of 8 years from HIV diagnosis to death. More than three quarters of our sample were Black Americans (80.8%) with men accounting or almost two-thirds (65.1%) and heterosexual transmission identified in more than half of the cases (53.4%). Men who have sex with men (MSM) risk behaviour and IDU accounted for 31.8% with 19.6% and 12.2%, respectively. During the year prior to death, approximately half (52.9%) of persons were tobacco smokers and one-fifth had excess alcohol use (20.5%) and non-IDU (20.7%). The median CD4 nadir was 51.0 cells/mm3 and the median last CD4 count prior to death was 106.0 cells/mm3. Less than
half (42.7%) of persons were on ART at the time of death. More than half of persons (60.2%) were CDC HIV Stage C. The most common underlying medical conditions seen were hypertension in 47.8% of persons, followed by hepatitis C (29.4%), depression (27.4%) and CKD (24.4%). Almost a quarter of the individuals were obese (8%) or overweight (16.1%) within one year of their death (Table 2). More than a quarter (25.9%) of the deaths was sudden and 15.2% unexpected. Tables 3 and 4 summarise the specific immediate causes of death and the presence of underlying or contributing factors, respectively. Nearly a third of the deaths were directly due to AIDSrelated infections (17.4%), AIDS-related malignancy (2%), or other infection (13.1%). Approximately, 47.5% of the deaths were due to non-AIDS causes. One-third of persons (33.1%) had an unidentified immediate, specific cause of death. AIDS infection (82.8%) was the most prevalent contributing or underlying condition leading to death followed by nonischemic heart or vascular disease (22.7%) and kidney disease (21.1%). AIDS-related immediate causes of death decreased from 36.6% in 2005 to 7.7% in 2013. A CDC stage C HIV condition or Hodgkin’s lymphoma was determined to be an underlying or contributing factor to death in 45% of persons. In an additional 77 (12.4%) patients who did not have a CDC stage C HIV condition or Hodgkin’s Lymphoma, it was determined, by applying CoDe guidelines, that immunodeficiency possibly or definitely contributed to death. Table 5 shows the result of logistic regression model showing factors associated with AIDS-related CoD. The model fits the data well with an adjusted r2 ¼ 0.147, indicating that the variables in the model explain approximately 15% of the variance of an AIDS-related death (ARD) outcome. Non-ARD was associated with a history of IDU (OR 8.80; 95% CI 1.103–70.262); CD4 nadir> ¼ 200 cells/mm3 (OR 4.96; 95% CI 1.399–17.563; VL < 400 copies/ml at death (OR 2.95; 95% CI 1.386–6.289); and being on ART at death (OR 2.02; 95% CI 1.074–3.808) There was no statistically significant association with the CoD and other factors such as history of tobacco use, excess alcohol use, non-IDU, race, CD4 prior to death, year of death, age at death and gender. In assessing trends over time, a fractional polynomial with regression identified the best model fit (m ¼ 2) based on a total of (n ¼ 417) death date values. This model indicated that there is a statistically significant trend over time in the change of last viral load prior to death, F(2,414) ¼ 48.95, p ¼ 0.001.31 This was confirmed by comparing median viral loads prior to death by year showing that in 2005 the median viral load was 71,450 decreasing to 31 in 2013.
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
4
International Journal of STD & AIDS 0(0) Table 1. History characteristics of persons with HIV and a date of death between 2005 and 2013. Characteristics
Frequency (%)
Number of deceased, n Median age at death (years) Median # Years HIV positive Gender, n (%) Men Women Missing HIV risk group, n (%) MSM IDU Heterosexual Transfusion Other Race, n (%) White American Black American Other Missing Habits, n (%) Tobacco smoking Excess alcohol use Non-injection drug use Injection drug use CD4 Nadir (mm3), median CD4 Prior to death (mm3), median Antiretroviral therapy at death, n (%) CDC HIV stage A B C Missing
459 48.0 (Q1 ¼ 40.0, Q3 ¼ 55.0) 8.0 (Q1 ¼ 4.0, Q3 ¼ 13.0) 299 (65.1) 149 (32.5) 11 (2.4) 90 (19.6) 56 (12.2) 245 (53.4) 12 (2.6) 19 (4.1) 65 (14.2) 371 (80.8) 6 (1.3) 17 (3.7) 243 (52.9) 94 (20.5) 95 (20.7) 32 (7.0) 51.0 (Q1 ¼ 9.0, Q3 ¼ 188.0) 106.0 (Q1 ¼ 18.0, Q3 ¼ 328.0) 196 (42.7) 85 (18.5) 44 (9.6) 277 (60.2) 52 (11.3)
Note: Percentages may not total to 100 due to missing data.
Fractional polynomial regression for CD4 count over time identified a relationship between the last CD4 count (median) and date of death. There was a statistically significant trend observed with CD4 prior to death increasing from 16 in 2004 to 360 in 2013 F (2,430) ¼ 11.45, p .001. Other time trends were assessed with no statistically significant findings. Demographic characteristics, such as HIV risk group, race, gender, age at death and on ART at death, were considered, but no statistically significant trends were observed. Regression analyses, when stratified by race, found associations between certain comorbid conditions and CoD. The statistically signifiant comorbidities by race
included hypertension (Black Americans p ¼ 0.001), diabetes (Black Americans p ¼ 0.029, White Americans p ¼ 0.040), dyslipidemia (Black Americans p ¼ 0.050), prior cardiovascular conditions (Black Americans p ¼ 0.001) and chronic hepatitis C (Black Americans p ¼ 0.007).
Discussion Data collected between 2005 and 2013 on 459 persons with HIV who died identified this group to be primarily men (65.1%), non-Hispanic, Black American (80.8%) and identified heterosexual transmission as the primary HIV risk factor (53.4%) with a median age at death of
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
Cima et al.
5
Table 2. Frequency histories, comorbid conditions in deceased HIV-infected persons, 2005–2013.
Table 4. Presence of underlying conditions contributing to causes of death in patients with HIV.
Comorbid conditions
Frequency (%)
Underlying conditions
Frequency (%)
Hypertension Hepatitis C infection Depression CKD Stage 3–5 Cardiovascular disease History of malignancy Diabetes Elevated liver function tests Dyslipidemia Hepatitis B infection Liver decompensation Psychosis Liver failure (before death) AIDS malignancy
220 135 126 112 92 85 72 69 61 47 41 37 37 35
AIDS infection AIDS malignancy Other infection Chronic viral Hepatitis Malignancy Diabetes Myocardial infarction Stroke Liver failure Renal failure Injury Central nervous system disease Heart or vascular Respiratory Unknown Total
380 14 46 71 38 27 26 17 21 97 74 28 104 20 29 992
(47.8) (29.4) (27.4) (24.4) (20.0) (18.5) (15.7) (15.0) (13.3) (10.2) (8.9) (8.0) (8.0) (7.6)
Table 3. Frequencies of immediate causes of death in deceased HIV-infected persons, 2005–2013. Immediate cause of death
Frequency (%)
AIDS infection AIDS malignancy Other infection Malignancy Myocardial infarction Stroke Liver failure Renal failure Injury Central nervous system disease Heart or vascular Respiratory Other Unknown Total
80 9 60 23 29 7 9 13 9 14 10 25 19 152 459
(17.4) (2.0) (13.1) (5.0) (6.3) (1.5) (2.0) (2.8) (2.0) (3.1) (2.2) (5.4) (4.1) (33.1) (100.0)
48 years and 8 years since HIV diagnosis. These data portray a common representation of the HIV epidemic in the Southern USA.20,32 Compared to other regions in the country which experience more reported MSM, younger and persons with longer time from diagnosis to death, these findings indicate a distinct pattern in keeping with research on similar populations. The health-related risk factors identified in our sample were 52.9% of persons smoked tobacco, 20.5% had a history of excess alcohol use, an equal percentage of non-IDU drug use and 7.0% with
(82.8) (3.1) (10.0) (15.5) (8.3) (5.9) (5.7) (3.7) (4.6) (21.1) (16.1) (6.1) (22.7) (4.4) (6.3)
Note: Conditions with n < 10: Pancreatitis (1), lactic acidosis (2), Gastrointestinal hemorrhage (2), primary pulmonary hypertension (7), lung embolus (2), chronic obstructive lung disease (7), endocrine disease (6), digestive system disease (5), obstetric complications (2), other (8).
reported IDU. Compared to other regions in the USA, the IDU percentage is somewhat less but representative of research on similar regions.33 The CD4 nadir median was 51 with 60.2% of persons diagnosed in CDC HIV Stage C. While these numbers are low for initial diagnosis, it is not uncommon in the South where persons experiencing more advanced disease present for care.34 Comorbid conditions correlated with health-related conditions observed among non-HIV infected persons, including hypertension, depression, diabetes and cardiovascular disease among others.35 While AIDSrelated diseases were the most frequently identified cause of death, it is important to note that it only accounted for 19.4% of all deaths as the primary CoD. Although using the CoDe algorithm, CDC stage C condition or immunodeficiency possibly or definitely contributed to death in 57.4%. In addition, AIDS was felt to be an underlying or contributing factor to death in the majority of individuals (82.8%). A larger number of persons fell into an unknown cause of death (33.1). Myocardial infarction and heart or vascular disease accounted for 8.5% (28.4% underlying) with respiratory disease accounting for 5.4% of immediate CoD. Current research suggests that the southern states such as SC may not be seeing the noted shift from AIDS-related to non-AIDS-related CoD in
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
6
International Journal of STD & AIDS 0(0)
Table 5. Patient characteristic associations with a non-aids related cause of death (n ¼ 221).a Variable
Unadj. OR
Adj. OR
p-Value
95% CI
Injection drug user vs no injection drug use CD4 Nadir > ¼ 200 cell/mm3 vs 400 copies/ml ART at time of death (Yes vs No) Bayesian Information Criterion ¼ 263.0
8.2 5.6 3.6 2.4
8.8 5.0 3.0 2.0
0.04 0.01
(STD)
[1–8] [PREPRINTER stage]
Int J STD AIDS OnlineFirst, published on February 17, 2015 as doi:10.1177/0956462415571970
Original research article
Cause of death in HIV-infected patients in South Carolina (2005–2013)
International Journal of STD & AIDS 0(0) 1–8 ! The Author(s) 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/0956462415571970 std.sagepub.com
Michael Cima1, R. David Parker1, Yasir Ahmed2, Sean Cook3, Shana Dykema4, Kristina Dukes3, Stephan Albrecht3 and Sharon Weissman3
Abstract The life span of persons with HIV has been greatly extended over the past 30 years due to novel therapies. In the developed world and urban settings, this results in a lifespan rivaling the lifespan of a person without HIV. A retrospective study was conducted on 459 patients of an urban, academic medical center who died between 2005 and 2013 in a medium-sized US city. Using the established Cause of Death Project (CoDe) protocol, we measured multiple factors including comorbidities, risk behaviours, contributing and underlying causes of death. This study is one of the few USbased studies using this validated protocol. Among the deaths, 25.9% were sudden and 15.2% were unexpected. Almost one-fifth were related to AIDS-related infections; 47.5% related to non-AIDS causes; with the remainder unknown. Statistically significant increases in CD4 counts and decreasing viral loads were observed over the study period. There were no statistically significant differences observed by HIV risk behaviour, race, gender, age at death, or on antiretorivirals at death. In support of the existing literature, improved HIV management appears to reduce the AIDS-related attributable death among patients observed in this study.
Keywords HIV, AIDS, comorbidity, mortality, cause of death Date received: 13 October 2014; accepted: 18 January 2015
Introduction With the widespread use of antiretroviral therapy (ART), HIV-associated mortality has dramatically declined in the USA and other developed nations.1,2 In the USA, life expectancy for newly diagnosed HIV positive patients increased from 10.5 years to 22.5 years between the years 1996 and 2005.3 A report from the North American AIDS Cohort Collaboration on Research and Design (NA_ACCORD) found that HIV-positive individuals treated in their twenties have a life expectancy into the early 1970’s.4 Numerous other studies support the finding that early initiation of HIV therapy correlates longer life expectancy in HIV positive individuals, approaching life expectancy of other chronic, non-HIV illnesses.5–8 As life expectancy of patients infected with HIV increases, comorbid (non-HIV/AIDS) conditions are more instrumental in the morbidity and mortality of HIV-positive patients who are in clinical care.9–12
Complications of aging are becoming more prevalent. HIV patients are at greater risk for aging complications such as osteoporosis, cancer, smoking-related illness and cardiovascular disease.13,14 In addition, ART therapy may accelerate or exacerbate some of these complications.15,16
1
School of Public Health, West Virginia University, Morgantown, WV, USA 2 Department of Internal Medicine, Texas Tech University, Odessa, TX, USA 3 Department of Medicine, University of South Carolina, Columbia, SC, USA 4 South Carolina Hospital Association, Columbia, SC, USA Corresponding author: Sharon Weissman, University of South Carolina, Department of Medicine, 2 Medical Park, Suite 205, Columbia, SC 29203, USA. Email: [email protected]
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
2
International Journal of STD & AIDS 0(0)
Several studies have shown that an increasing proportion of deaths in HIV-infected patients are attributable to causes not traditionally considered to be related to HIV.9–12,17,18 In several of these studies, the nonHIV/AIDS conditions account for greater than 50% of the deaths.9,10,12 These findings are supported by large cohort studies showing these non-HIV/AIDS contribute significantly to mortality among persons with HIV.17,18 This shift from AIDS-related to non-HIV/ AIDS causes of death (CoD) is most prominent in patients receiving ART and with higher current and initial (nadir) CD4 counts.9,10,12,17,18 Thus a population, who is not receiving the maximum benefit from ART therapy, either because of late stage HIV diagnosis or limited access to care, may not experience this same shift in CoD. Therefore, persons in areas of the USA who are more traditionally impacted by barriers in accessing care and corollaries, such as poverty, homelessness, lack of insurance, substance misuse and mental health issues are most likely at risk, such as persons in the Southeastern states.19,20 Multiple CDC MMWR reports have shown that a large portion of HIV-infected patients in South Carolina (SC) are diagnosed late in their course of HIV infection.21,22 From 2001 to 2005, 41% of newly diagnosed HIV patients in SC were late testers, presenting with an AIDS diagnosis or developed AIDS within one year of their HIV diagnosis.23 Furthermore, in SC a significant proportion of people living with HIV and AIDS live in rural areas where access to care and HIV testing is often a barrier.24–27 Thus, SC might not see the same shift to non-HIV/AIDS CoD. There are no published data on CoD for HIV-infected patients neither in the southern states nor in SC. This information is important for targeting future prevention, treatment efforts and research efforts. The aim of this study is to determine CoD in HIVinfected individuals in SC. The secondary aims are to explore the relationships between various factors, such as patient characteristics, immunologic and viral logic parameters, and high risk behaviours such as drug misuse and cause of death. Lastly, to determine if there are any trends over time in CoD or patient characteristics.
This hospital and clinic provide approximately 50% of all HIV services in the state and 90% of HIV care in an eight county area. Patient care and billing databases were queried to generate a patient list for outpatient care. Inpatient data were collected using the electronic health record. To increase case finding, especially among patients lost to care, a list of patients who were last seen on or before 1 January 2008 was compared to state and national death registries. The Cause of Death Project (CoDe) protocol was employed for the purposes of this study.28 The CoDe project is a multinational collaborative effort to develop a standardised approach to data collection and determination of CoD in HIV-infected patients. The protocol of the original study is publicly available and can be used free of charge. Included in the protocol is a standardised Case Report Form (CRF) and algorithm for determining the cause of death. Patient medical records were reviewed and data on HIV treatment, complications, CD4 counts, HIV viral load, comorbid conditions and cause of death were extracted using the CRF. Additional data, not found on the CoDe CRF, were extracted on HIV risk factors, race, substance use, chronic kidney disease (CKD), malignancy, obesity and ART use through a supplemental data collection form.
Classification of cause of death In accordance with CoDe protocol, two clinicians reviewed various data sources to determine a specific cause of death. If a consensus was reached, then CoD was established. If there was disagreement between these two reviewers, or both coded the CoD as unknown or unclassifiable, the specific case was referred to one or more additional reviewers to arbitrate. If a consensus was still not reached, the case was classified according to majority decision or deemed unknown. As stated in the protocol, autopsy results were used to determine CoD when available. In situations where autopsy results were not available, the cause of death was determined based on review of medical records, interview with patients’ primary medical provider, death certificate, obituary, registry, or family report of cause. Family members were not contacted by study personnel.
Methods Sample and data collection
Data analysis
A retrospective medical record review collected data on patients who died between 1 January 2005 and 31 December 2013. CoD data on HIV-infected patients who received inpatient care and/or outpatient care from an infectious diseases clinic and associated hospital in a medium-sized, Southern city were collected.
To determine unadjusted associations between variables, univariate analyses including Fisher’s Exact Test for categorical data and t-tests or the nonparametric equivalent were used. Non-parametrics were used for more exact tests in instances when the cell sizes were too small to allow standard parametric
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
Cima et al.
3
analyses or in situations where there was a high degree of data variance. Different multivariable regressions determined the presence and statistical strength of relationships in models designed to best fit the data. For analyses which included primarily categorical data with a dichotomous outcome, multivariable logistic regressions were used. Multivariable models allowed analyses to consider the presence and effect of multiple variables within the model. Variables included in these models were: smoking tobacco, excessive alcohol consumption, non-injection illicit drug use (non-IDU), illicit injection drug use (IDU), race, gender, last CD4 count, CD4 nadir, viral load prior to death, age at death and ART at death. Stepwise model reduction used a statistically significant criterion of (p ¼ 0.05) for individual variable removal from the model. However, also included were considerations in the variance, Bayesian Information Criteria (BIC) and effect sizes in conjunction with the law of parsimony determining the best model. Additionally, given the potential for differentiation between clinical and statistical significance in any biomedical model, the investigators ensured that the interactions between variables were measured (as indicated) as well as confounding, effect modification and collinearity. In the case of differences between statistically significant and clinically significant models, a determination was made at that time to determine the best model for fit. R square and adjusted r-square values helped to inform these decisions, as well as the confidence intervals, effect sizes and the clinical knowledge of the investigators as well as the aforementioned methods for goodness of fit of each model to the data. Fractional polynomials were used in conjunction with the appropriate regression were used for continuous variables to determine trends across study years.29,30
Results The characteristics of the sample are presented in Tables 1 and 2. Our total sample size (n ¼ 459) had a median age of 48 years at death with a median of 8 years from HIV diagnosis to death. More than three quarters of our sample were Black Americans (80.8%) with men accounting or almost two-thirds (65.1%) and heterosexual transmission identified in more than half of the cases (53.4%). Men who have sex with men (MSM) risk behaviour and IDU accounted for 31.8% with 19.6% and 12.2%, respectively. During the year prior to death, approximately half (52.9%) of persons were tobacco smokers and one-fifth had excess alcohol use (20.5%) and non-IDU (20.7%). The median CD4 nadir was 51.0 cells/mm3 and the median last CD4 count prior to death was 106.0 cells/mm3. Less than
half (42.7%) of persons were on ART at the time of death. More than half of persons (60.2%) were CDC HIV Stage C. The most common underlying medical conditions seen were hypertension in 47.8% of persons, followed by hepatitis C (29.4%), depression (27.4%) and CKD (24.4%). Almost a quarter of the individuals were obese (8%) or overweight (16.1%) within one year of their death (Table 2). More than a quarter (25.9%) of the deaths was sudden and 15.2% unexpected. Tables 3 and 4 summarise the specific immediate causes of death and the presence of underlying or contributing factors, respectively. Nearly a third of the deaths were directly due to AIDSrelated infections (17.4%), AIDS-related malignancy (2%), or other infection (13.1%). Approximately, 47.5% of the deaths were due to non-AIDS causes. One-third of persons (33.1%) had an unidentified immediate, specific cause of death. AIDS infection (82.8%) was the most prevalent contributing or underlying condition leading to death followed by nonischemic heart or vascular disease (22.7%) and kidney disease (21.1%). AIDS-related immediate causes of death decreased from 36.6% in 2005 to 7.7% in 2013. A CDC stage C HIV condition or Hodgkin’s lymphoma was determined to be an underlying or contributing factor to death in 45% of persons. In an additional 77 (12.4%) patients who did not have a CDC stage C HIV condition or Hodgkin’s Lymphoma, it was determined, by applying CoDe guidelines, that immunodeficiency possibly or definitely contributed to death. Table 5 shows the result of logistic regression model showing factors associated with AIDS-related CoD. The model fits the data well with an adjusted r2 ¼ 0.147, indicating that the variables in the model explain approximately 15% of the variance of an AIDS-related death (ARD) outcome. Non-ARD was associated with a history of IDU (OR 8.80; 95% CI 1.103–70.262); CD4 nadir> ¼ 200 cells/mm3 (OR 4.96; 95% CI 1.399–17.563; VL < 400 copies/ml at death (OR 2.95; 95% CI 1.386–6.289); and being on ART at death (OR 2.02; 95% CI 1.074–3.808) There was no statistically significant association with the CoD and other factors such as history of tobacco use, excess alcohol use, non-IDU, race, CD4 prior to death, year of death, age at death and gender. In assessing trends over time, a fractional polynomial with regression identified the best model fit (m ¼ 2) based on a total of (n ¼ 417) death date values. This model indicated that there is a statistically significant trend over time in the change of last viral load prior to death, F(2,414) ¼ 48.95, p ¼ 0.001.31 This was confirmed by comparing median viral loads prior to death by year showing that in 2005 the median viral load was 71,450 decreasing to 31 in 2013.
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
4
International Journal of STD & AIDS 0(0) Table 1. History characteristics of persons with HIV and a date of death between 2005 and 2013. Characteristics
Frequency (%)
Number of deceased, n Median age at death (years) Median # Years HIV positive Gender, n (%) Men Women Missing HIV risk group, n (%) MSM IDU Heterosexual Transfusion Other Race, n (%) White American Black American Other Missing Habits, n (%) Tobacco smoking Excess alcohol use Non-injection drug use Injection drug use CD4 Nadir (mm3), median CD4 Prior to death (mm3), median Antiretroviral therapy at death, n (%) CDC HIV stage A B C Missing
459 48.0 (Q1 ¼ 40.0, Q3 ¼ 55.0) 8.0 (Q1 ¼ 4.0, Q3 ¼ 13.0) 299 (65.1) 149 (32.5) 11 (2.4) 90 (19.6) 56 (12.2) 245 (53.4) 12 (2.6) 19 (4.1) 65 (14.2) 371 (80.8) 6 (1.3) 17 (3.7) 243 (52.9) 94 (20.5) 95 (20.7) 32 (7.0) 51.0 (Q1 ¼ 9.0, Q3 ¼ 188.0) 106.0 (Q1 ¼ 18.0, Q3 ¼ 328.0) 196 (42.7) 85 (18.5) 44 (9.6) 277 (60.2) 52 (11.3)
Note: Percentages may not total to 100 due to missing data.
Fractional polynomial regression for CD4 count over time identified a relationship between the last CD4 count (median) and date of death. There was a statistically significant trend observed with CD4 prior to death increasing from 16 in 2004 to 360 in 2013 F (2,430) ¼ 11.45, p .001. Other time trends were assessed with no statistically significant findings. Demographic characteristics, such as HIV risk group, race, gender, age at death and on ART at death, were considered, but no statistically significant trends were observed. Regression analyses, when stratified by race, found associations between certain comorbid conditions and CoD. The statistically signifiant comorbidities by race
included hypertension (Black Americans p ¼ 0.001), diabetes (Black Americans p ¼ 0.029, White Americans p ¼ 0.040), dyslipidemia (Black Americans p ¼ 0.050), prior cardiovascular conditions (Black Americans p ¼ 0.001) and chronic hepatitis C (Black Americans p ¼ 0.007).
Discussion Data collected between 2005 and 2013 on 459 persons with HIV who died identified this group to be primarily men (65.1%), non-Hispanic, Black American (80.8%) and identified heterosexual transmission as the primary HIV risk factor (53.4%) with a median age at death of
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
Cima et al.
5
Table 2. Frequency histories, comorbid conditions in deceased HIV-infected persons, 2005–2013.
Table 4. Presence of underlying conditions contributing to causes of death in patients with HIV.
Comorbid conditions
Frequency (%)
Underlying conditions
Frequency (%)
Hypertension Hepatitis C infection Depression CKD Stage 3–5 Cardiovascular disease History of malignancy Diabetes Elevated liver function tests Dyslipidemia Hepatitis B infection Liver decompensation Psychosis Liver failure (before death) AIDS malignancy
220 135 126 112 92 85 72 69 61 47 41 37 37 35
AIDS infection AIDS malignancy Other infection Chronic viral Hepatitis Malignancy Diabetes Myocardial infarction Stroke Liver failure Renal failure Injury Central nervous system disease Heart or vascular Respiratory Unknown Total
380 14 46 71 38 27 26 17 21 97 74 28 104 20 29 992
(47.8) (29.4) (27.4) (24.4) (20.0) (18.5) (15.7) (15.0) (13.3) (10.2) (8.9) (8.0) (8.0) (7.6)
Table 3. Frequencies of immediate causes of death in deceased HIV-infected persons, 2005–2013. Immediate cause of death
Frequency (%)
AIDS infection AIDS malignancy Other infection Malignancy Myocardial infarction Stroke Liver failure Renal failure Injury Central nervous system disease Heart or vascular Respiratory Other Unknown Total
80 9 60 23 29 7 9 13 9 14 10 25 19 152 459
(17.4) (2.0) (13.1) (5.0) (6.3) (1.5) (2.0) (2.8) (2.0) (3.1) (2.2) (5.4) (4.1) (33.1) (100.0)
48 years and 8 years since HIV diagnosis. These data portray a common representation of the HIV epidemic in the Southern USA.20,32 Compared to other regions in the country which experience more reported MSM, younger and persons with longer time from diagnosis to death, these findings indicate a distinct pattern in keeping with research on similar populations. The health-related risk factors identified in our sample were 52.9% of persons smoked tobacco, 20.5% had a history of excess alcohol use, an equal percentage of non-IDU drug use and 7.0% with
(82.8) (3.1) (10.0) (15.5) (8.3) (5.9) (5.7) (3.7) (4.6) (21.1) (16.1) (6.1) (22.7) (4.4) (6.3)
Note: Conditions with n < 10: Pancreatitis (1), lactic acidosis (2), Gastrointestinal hemorrhage (2), primary pulmonary hypertension (7), lung embolus (2), chronic obstructive lung disease (7), endocrine disease (6), digestive system disease (5), obstetric complications (2), other (8).
reported IDU. Compared to other regions in the USA, the IDU percentage is somewhat less but representative of research on similar regions.33 The CD4 nadir median was 51 with 60.2% of persons diagnosed in CDC HIV Stage C. While these numbers are low for initial diagnosis, it is not uncommon in the South where persons experiencing more advanced disease present for care.34 Comorbid conditions correlated with health-related conditions observed among non-HIV infected persons, including hypertension, depression, diabetes and cardiovascular disease among others.35 While AIDSrelated diseases were the most frequently identified cause of death, it is important to note that it only accounted for 19.4% of all deaths as the primary CoD. Although using the CoDe algorithm, CDC stage C condition or immunodeficiency possibly or definitely contributed to death in 57.4%. In addition, AIDS was felt to be an underlying or contributing factor to death in the majority of individuals (82.8%). A larger number of persons fell into an unknown cause of death (33.1). Myocardial infarction and heart or vascular disease accounted for 8.5% (28.4% underlying) with respiratory disease accounting for 5.4% of immediate CoD. Current research suggests that the southern states such as SC may not be seeing the noted shift from AIDS-related to non-AIDS-related CoD in
Downloaded from std.sagepub.com at UCSF LIBRARY & CKM on February 24, 2015
XML Template (2015) [16.2.2015–2:15pm] //blrnas3.glyph.com/cenpro/ApplicationFiles/Journals/SAGE/3B2/STDJ/Vol00000/150013/APPFile/SG-STDJ150013.3d
(STD)
[1–8] [PREPRINTER stage]
6
International Journal of STD & AIDS 0(0)
Table 5. Patient characteristic associations with a non-aids related cause of death (n ¼ 221).a Variable
Unadj. OR
Adj. OR
p-Value
95% CI
Injection drug user vs no injection drug use CD4 Nadir > ¼ 200 cell/mm3 vs 400 copies/ml ART at time of death (Yes vs No) Bayesian Information Criterion ¼ 263.0
8.2 5.6 3.6 2.4
8.8 5.0 3.0 2.0
0.04 0.01
