Are we missing opioid-related deaths among people with HIV?

Abstract

Background:

Ascertainment of unnatural and overdose death may be unreliable among individuals with life-limiting conditions such as HIV infection. We sought to determine whether the relationship between opioid use and unnatural death differs among decedents with HIV (DWH) and those without.

Methods:

Decedents in the Veterans Aging Cohort Study (VACS) from 2002 to 14 were linked to the National Death Index cause of death file. Deaths were classified as unnatural, overdose (a subset of unnatural), or other. We defined opioid use as self-reported illicit use or receipt of prescribed opioids. Treating unnatural and overdose deaths as outcomes, we calculated odds ratios for opioid exposure by HIV status, with and without adjustment for disease severity using VACS Index.

Results:

Among 561 decedents without HIV (DWOH) and 884 DWH, 11 % and 8 % respectively were classified as unnatural deaths and 4 % and 2 % were classified as overdose deaths. Among DWOH, opioid use was associated with 2-fold greater odds of unnatural (OR 2.3; 95 % CI 1.3–4.0) and 4-fold greater odds of overdose death (OR 4.5; 95 % CI 1.5–13.7); in adjusted analyses, opioid use was associated with unnatural death (OR 2.6; 95 % CI 1.3–4.9) and with overdose (OR 4.2; 95 % CI 1.4–12.7). Opioid use was not associated with unnatural or overdose death among DWH.

Conclusion:

Opioid use was strongly associated with unnatural and overdose death among DWOH but not among DWH suggesting potential differential misclassification. Caution should be used in interpreting prevalence, incidence and risk factors for unnatural and overdose cause of death among patients with life-limiting conditions such as HIV.

1. Introduction

In light of the continued rise in opioid-related deaths internationally (Dart et al., 2015; Martins et al., 2015; O’Donnell et al., 2017), the substantially improved life expectancy of people with HIV infection, (PWH) (Croxford et al., 2017; Smith et al., 2014; Trickey et al., 2017) and common use of opioids among PWH (Becker et al., 2016; Edelman et al., 2013), accurately identifying opioid-related deaths among PWH is increasingly important. While cause of death attribution is known to be challenging (Crow et al., 2009; Hoffman et al., 2003), most consider differentiation between unnatural death (including overdose) and natural death (including death from HIV) straightforward. However, it may be problematic in the face of a potentially life-limiting disease such as HIV infection.

Opioid-related unnatural deaths may take multiple forms. While overdose is the most often discussed, other unnatural causes – e.g. motor vehicle collisions – have well-established links to opioid use and also appear to be increasing in the opioid crisis era (Gomes et al., 2013). Conversely, even disease related – so called “natural”– deaths may be precipitated by opioid use, as opioids increase risk of pneumonia (Edelman et al., 2019) and other conditions, such as falls (Krebs et al., 2016) and fractures (Saunders et al., 2010), that can have life-threatening sequelae.

Recognizing that assigning cause of death is challenging and may be subject to misclassification in individuals with a life-limiting condition (Justice, 2009), we sought to examine whether the association between opioid use (prescribed and illicit) and unnatural death (including overdose) varied by HIV status, after controlling for severity of illness using the Veterans Aging Cohort Study (VACS) Index, a validated prognostic index (Justice et al., 2006).

2. Methods

This study was approved by the human subjects subcommittees of Veterans Affairs (VA) Connecticut Healthcare System and Yale University. Our data source was the VACS survey sample, a cohort of PWH and uninfected veterans matched 2:1 on age, gender, race and site of care. Participants complete surveys at enrollment and during ongoing follow-up on alcohol and drug use and other health behaviors; survey data is linked to the electronic health record (EHR), e.g. pharmacy data. Our analytic sample consisted of VACS decedents from 2002 to 2014 who responded to at least one follow-up survey.

2.1. Cause of death

We linked VACS records with the National Death Index cause of death file, which codes underlying and contributory causes of death from information provided by whoever certified the death. Because medical examiners or coroners typically only certify deaths when there is trauma, suspicious circumstances or uncertainty, the certifier is usually a treating physician (Brooks and Reed, 2015). Based on International Classification of Diseases (ICD) 9 and/or 10 codes, we categorized cause of death, either underlying or contributory, into either 1) unnatural death, i.e. causes of death that could be directly opioid-related such as motor-vehicle crash and overdose, and 2) all other causes of death, both HIV-related and not. As described below, in some analyses we examined overdose deaths as a subset of unnatural deaths, and excluded HIV-related deaths for greater specificity.

2.2. Illicit and prescribed opioid use

We defined three different measures of opioid use to improve sensitivity and validity; two based on VACS survey self-report of illicit opioid use (heroin use and non-medical use of prescription opioids), and one from EHR pharmacy data (receipt of prescribed opioids). A positive response on either self-report item—defined as any use in the past year—on any follow-up survey was categorized as opioid use. From the EHR, we ascertained outpatient oral or transdermal prescription opioid receipt during the year prior to death using our published methodology (Edelman et al., 2013). Decedents with ≥ 90 days of pharmacy opioid receipt were categorized as having opioid use; we used the 90-day threshold because shorter supplies of opioids are very common in the sample.

2.3. Demographic and clinical characteristics

We included year of study enrollment; age at enrollment, categorized as < 50, 50−64, ≥ 65; race/ethnicity; and sex. Clinical characteristics included presence of chronic hepatitis C infection (HCV), based on ICD9/10 code, positive antibody or detectable RNA as a marker of likely history of injection drug use; and VACS Index 2.0 (Tate et al., 2019), a well-validated prognostic score composed of age, hemoglobin, fibrosis-4, eGFR, HCV status, white blood cell count, BMI, albumin, CD4, and viral load. Race is included in the estimation of FIB-4 and eGFR.

2.4. Statistical analyses

We first described the sample using frequencies/proportions and then compared frequencies/proportions between decedents without HIV (DWOH) and with HIV (DWH) using chi-squared tests. We calculated the crude odds ratio of unnatural death and overdose, for each opioid exposure measure, stratified by decedent’s HIV status. Next, we performed two logistic regressions stratified by HIV status, one modeling unnatural death and the second modeling overdose death, using any opioid use as the primary exposure of interest, adjusting for VACS Index 2.0 score by five-point increments. Finally, we performed two sensitivity analyses: repeating the logistic regression models 1) restricting the sample to decedents with a prognostic score < 100, as individuals with scores ≥ 100 have a 20 % 1-year mortality risk (Justice et al., 2013) and 2) excluding decedents with HIV-related deaths. The sensitivity analyses sought to 1) remove individuals with high likelihood of death from non-opioid-related causes and 2) examine whether HIV status obscured the relationship between opioid exposure and cause of death even among DWH who did not die of HIV-related causes.

3. Results

The sample (N = 1445) consisted of 561 DWOH and 884 DWH (Table 1). DWH were more likely to be < 50 years old, black, and to have HCV, and were less likely to be ≥ 65 years old or Hispanic, compared to DWOH. Any opioid use (i.e. self-report or prescribed) was equally likely among DWOH (42 %) and DWH (44 %); (p = 0.4). Neither of the individual measures of opioid use differed by HIV status.

Table 1.

Characteristics of 1445 decedents, Veterans Aging Cohort Study.

	Overall (N = 1445)	Decedents without HIV (N = 561)	Decedents with HIV (N = 884)	p value
Year enrolled, median (interquartile range)	2003 (2002, 2003)	2003 (2002, 2003)	2003 (2002,	0.03
Age at enrollment n (%)				< 0.001
< 50	485 (34)	138 (25)	347 (39)
50−64	771 (53)	305 (54)	466 (53)
65+	189 (13)	118 (21)	71 (8)
Race				0.0003
White, non-Hispanic	360 (25)	160 (29)	200 (23)
Black, non-Hispanic	955 (66)	335 (60)	620 (70)
Hispanic	114 (8)	59 (11)	55 (6)
Other/unknown	16 (1)	7 (1)	9 (1)
Sex				0.77
Female	19 (1)	8 (1)	11 (1)
HCV	619 (43)	185 (33)	434 (49)	< 0.001
Opioid use
Illicit opioids including heroin	497 (34)	187 (33)	310 (35)	0.50
Non-medical use of prescription opioids	322 (22)	125 (22)	197 (22)	1.00
Prescribed opioid receipt > = 90days	290 (20)	119 (21)	171 (19)	0.39
Any of above	624 (43)	234 (42)	390 (44)	0.37
Year died				< 0.001
2002−2009	571 (40)	166 (30)	405 (46)
2009−2014	874 (60)	395 (70)	479 (54)
Unnatural death	130 (9)	61 (11)	69 (8)	0.05
Overdose death	39 (3)	21 (4)	18 (2)	0.05

Among DWOH, 61 deaths (11 %) were classified as unnatural, 21 (4 %) of which were overdose. DWOH with opioid use (illicit or prescribed), relative to those without opioid use, were 2.3 (95 % CI 1.3–4.0) times as likely to be coded as unnatural death (p = 0.004) and 4.5 (95 % CI 1.5–13.7) times as likely to be coded as overdose death (p = 0.003) (Fig. 1). Among DWH, 331 deaths (37 %) were classified as HIV-related; 69 (8 %) as unnatural, of which 18 (2 %) were classified as overdose. However, among DWH, there was no difference in the proportion of unnatural death by opioid exposure status; odds ratios were not significant (1.1 [95 % CI 0.7–1.8] and 1.0 [95 % CI 0.4–2.4], respectively). This pattern – significant ratios comparing DWOH with and without opioid use – was consistent for unnatural death as a whole and for the more specific overdose death, for each type of opioid use (Fig. 1). In contrast, among DWH, there was not a significant difference observed between those with and without opioids for unnatural and overdose death across all measures of exposure. DWH with opioid use were no more likely to have death attributed to an opioid-related unnatural cause than DWH without opioid use.

Fig. 1.

Forest plot of crude odds ratios of types of opioid use and unnatural deaths and overdose, stratified by HIV status. DWOH = decedents without HIV; DWH = decedents with HIV; NMUPO = non-medical use of prescription opioids

Our findings persisted after adjustment for VACS Index 2.0 (Table 2) wherein opioid use was associated with all-cause unnatural death (OR 2.6; 95 % CI 1.3–4.9, p = 0.005) among DWOH but was not among DWH (OR 1.3; 95 % CI 0.8–2.3, p = 0.31). Similarly, opioid use was associated with overdose (OR 4.2; 95 % CI 1.4–12.7, p = 0.01) among DWOH but not associated with overdose (OR 1.4; 95 % CI 0.5–3.7, p = 0.55) among DWH. Our two sensitivity analyses–restricting the sample to 1) decedents with VACS Index 2.0 score < 100 and 2) decedents without HIV-related deaths – revealed similar findings (Supplementary material can be found by accessing the online version of this paper at https://nam05.safelinks.protection.outlook.com/?url=http%3A%2F%2Fdx.doi.org&data=02%7C01%7Cwilliam.becker%40yale.edu%7C06f25e23ad014d18877d08d79f895c79%7Cdd8cbebb21394df8b4114e3e87abeb5c%7C0%7C0%7C637153284626494431&sdata=KbzB8yfzljLJYCRvqdgO%2BdkOqcsyZ9kvj968wmEuSO4%3D&reserved=0). A post hoc sensitivity analysis stratifying by 2002−2009 and 2009−2014 was conducted after we observed imbalance in years of death and revealed similar findings (Supplementary material can be found by accessing the online version of this paper at https://nam05.safelinks.protection.outlook.com/?url=http%3A%2F%2Fdx.doi.org&data=02%7C01%7Cwilliam.becker%40yale.edu%7C06f25e23ad014d18877d08d79f895c79%7Cdd8cbebb21394df8b4114e3e87abeb5c%7C0%7C0%7C637153284626494431&sdata=KbzB8yfzljLJYCRvqdgO%2BdkOqcsyZ9kvj968wmEuSO4%3D&reserved=0).

Table 2.

Multivariable models of unnatural and overdose death, stratified by HIV status.

		Decedents without HIV, n = 489		Decedents with HIV, n = 811
Outcome	Parameter	AOR (95 % CI)	p value	AOR (95 % CI)	p value
Unnatural death	Any opioid use	2.55 (1.33, 4.87)	0.005	1.33 (0.77, 2.29)	0.31
	VACS Index 2.0 score (by 5 units)	0.69 (0.61, 0.77)	< 0.001	0.78 (0.73, 0.84)	< 0.001
Overdose	Any opioid use	4.15 (1.35, 12.73)	0.01	1.36 (0.49, 3.74)	0.55
	VACS index 2.0 score (by 5 units)	0.79 (0.68, 0.92)	0.002	0.81 (0.72, 0.92)	0.001

4. Discussion

In a large national sample of decedents with and without HIV, evidence of opioid use was associated with increased odds of unnatural death among DWOH but was not associated with opioid-related unnatural death among DWH, regardless of type of opioid use and controlling for severity of illness. That this relationship persisted after exclusion of HIV-related deaths and those with high VACS Index 2.0 scores further strengthens our findings.

It is reasonable to expect that opioid use (both illicit and prescribed) would increase the proportion of unnatural death and more explicitly overdose death. As expected, among DWOH we found a strong association using multiple measures of opioid exposure. It is concerning that this association was not seen among DWH.

There are several possible explanations. One possibility is that DWH died of “natural” causes before they had a chance to die from unnatural causes. This appears unlikely in light of adjustment for severity of illness, which was protective against unnatural and overdose death, and the sensitivity analyses excluding those with a high probability of death due to disease burden. Another possibility is that providers caring for those with HIV were either biased by the knowledge of HIV infection away from the signs of an unnatural death or did not consider the risks of unnatural death in this context (Carroll et al., 2019). A third possibility is that opioid exposure differentially hastened a “natural” death among those with HIV. This is also a possibility considering the increased risk of pneumonia associated with HIV (Edelman et al., 2013).

Nevertheless, these findings have important implications for future epidemiologic research on the association between opioid exposure and harm, among those with potentially life-limiting diseases such as HIV infection. Future work should consider whether other substance-related harm, for example from stimulants, is also potentially obscured by the presence of HIV infection. Our findings suggest that, barring improvement in accuracy of cause of death certification, all-cause mortality should be considered in studies examining opioid-related harm since the true rate of opioid-related unnatural deaths may be underestimated. This is in part because all-cause mortality encompasses not only unnatural causes of death that have a proximal and direct causal relationship with opioids, such as overdose, but also “natural” causes, where opioids may be contributory in a more gradual and/or indirect fashion; as appears to be the case with community-acquired pneumonia (Edelman et al., 2019; Wiese et al., 2018). A better understanding of the relationship between opioid use and all-cause mortality could improve the information with which providers and patients make treatment decisions.

Our study has important strengths. Mortality ascertainment within VACS is excellent (Justice et al., 2006), thereby ensuring the completeness of our request to the National Death Index for cause of death. In addition, our uninfected comparators are also veterans receiving care within the national VA healthcare system making them very much like those with HIV infection in our sample, as was illustrated by the equally high prevalence of prior opioid exposure in both groups. It also ensures that confounders including HCV, age and laboratory values were measured consistently across HIV status. Finally, by using a validated risk index (VACS Index 2.0), we were able to adjust for the probability of the death resulting from physiologic frailty.

Our study also has some limitations. First, it was performed in a single health system so may not be generalizable to other health systems; however, in this study, cause of death was frequently coded outside of the VA, suggesting wider generalizability of our findings. Second, we examined a predominantly male population and thus our results may not generalize to women. Third, our measures of opioid use were binary and (for two out of the three measures) self-reported, meaning we may have had imprecise estimates of the degree of opioid exposure. However, it is likely that the lack of precision would have been (on average) equal in both DWH and DWOH, thus minimizing bias in this aspect. We did not measure cannabis use; as medical cannabis became more prevalent in this time period and HIV a commonly listed indication for medical certification, it is possible that individuals with HIV were using fewer opioids due to effective pain control with cannabis.

5. Conclusions

We found evidence for potential under-attribution of unnatural and overdose death associated with opioid exposure among decedents with HIV infection. Given the overlapping epidemics of opioid use and HIV infection, this finding has important implications for not only epidemiologic research on opioid-related harms, but also but also policies resulting from cause of death data.