Individual, community, and structural factors associated with linkage to HIV care among people diagnosed with HIV in Tennessee.

OBJECTIVE: We assessed trends and identified individual- and county-level factors associated with individual linkage to HIV care in Tennessee (TN).
METHODS: TN residents diagnosed with HIV from 2012-2016 were included in the analysis (n = 3,751). Individuals were assigned county-level factors based on county of residence at the time of diagnosis. Linkage was defined by the first CD4 or HIV RNA test date after HIV diagnosis. We used modified Poisson regression to estimate probability of 30-day linkage to care at the individual-level and the contribution of individual and county-level factors to this outcome.
RESULTS: Both MSM (aRR 1.23, 95%CI 0.98-1.55) and women who reported heterosexual sex risk factors (aRR 1.39, 95%CI 1.18-1.65) were more likely to link to care within 30-days than heterosexual males. Non-Hispanic Black individuals had poorer linkage than White individuals (aRR 0.77, 95%CI 0.71-0.83). County-level mentally unhealthy days were negatively associated with linkage (aRR 0.63, 95%CI: 0.40-0.99).
CONCLUSIONS: Racial disparities in linkage to care persist at both individual and county levels, even when adjusting for county-level social determinants of health. These findings suggest a need for structural interventions to address both structural racism and mental health needs to improve linkage to care and minimize racial disparities in HIV outcomes.

Introduction

In 2018, despite representing only 38% of the US population, Southern states accounted for 46% of persons living with HIV (PLWH), and 52% of new diagnoses [1]. To gauge where the gaps in HIV care existed in the southeastern state of Tennessee (TN), the TN Department of Health (TDH) completed its first continuum of care analysis in 2010, revealing that TN under-performed relative to the general US population in both linking newly diagnosed PLWH to care within 90 days (66% vs. 80%) and retaining patients in HIV care over time (37% vs. 46%) [2, 3]. Since then, timely linkage to HIV care has been emphasized by the US Department of Health and Human Services’ as a key pillar in the country’s Ending the HIV Epidemic (EtE) Initiative, but TN has continued to lag in linkage to care indices [4, 5].

In addition, racial/ethnic disparities have persisted across the US over decades, with non-Hispanic Black (Black) individuals typically experience worse outcomes than other racial/ethnic groups across the entire continuum of HIV care [6]. Racial disparities have been variably attributed to higher rates of poverty, unemployment, and stigma–inequities even more pronounced in the Southern US–and might drive some of TN’s poor performance on linkage to HIV care [6-8]. There are limited data characterizing whether and to what extent significant racial disparities in HIV outcomes remain after accounting for both individual and county-level factors known to be associated with poor health outcomes, and disproportionately impacting racial minorities [7, 8]. Such studies are important in the Southern US, home to both a higher rate of incident HIV and more pronounced racial disparities in HIV-related health outcomes than other regions in the US [1, 9].

To improve performance along the HIV care continuum, TDH launched a number of initiatives between 2010 and 2015, including capacity building and infrastructure changes to improve the accuracy and efficiency of HIV testing and reporting, as well as the implementation of a social networking program for Black men who have sex with men (MSM) to address linkage and re-engagement in care [10, 11]. In the wake of these concerted efforts, the objective of this analysis was to integrate individual and county-level data assessing individual, community, and structural drivers of healthcare outcomes to understand 1) trends in linkage to HIV care in TN over time, 2) drivers of poor linkage to care outcomes and 3) drivers of ongoing racial disparities in these outcomes in TN and in counties with the highest HIV burden.

Materials and methods

Ethical approval

We obtained a waiver of consent and IRB approval from Vanderbilt University Medical Center (Protocol 173 no. 17119, Nashville, TN, USA), and TDH (protocol no. 1097644–4).

Study setting and design

We conducted a retrospective cohort study of persons who resided in TN and were newly diagnosed PLWH between January 1, 2012 and December 31, 2016. We assessed trends in linkage to care rates over the study period and individual, community, and structural predictors of linkage to HIV care (measured at the county level). The outcome of interest was linkage to HIV care, which was defined as receipt of the first CD4 or HIV-1 RNA test result captured via TN’s enhanced HIV/AIDS reporting system (eHARS) after the date of diagnosis, and was assessed at 30, 60, and 90 days.

Measures

Individual-level measures

Individual-level variables obtained from eHARS included: year of diagnosis, age at diagnosis (both validated via standardized data cleaning measures to account for repeat testing), sex, race/ethnicity (White/non-Hispanic (White), Black, Hispanic/all races, other/unknown), HIV risk factor (heterosexual contact, MSM, injection drug use (IDU; includes MSM/IDU), other, unknown), site of diagnosis (inpatient facility/emergency room (ER), outpatient facility, health department or sexually transmitted disease (STD)/family planning clinic, blood bank, correctional facility, other/unknown, missing) and ZIP Code of residence. ZIP Codes in which there were fewer than 5 HIV cases reported were suppressed according to TDH data suppression requirements. We combined sex and HIV risk factor into one variable with the following categories: male/heterosexual, male/MSM, male/IDU, male/other-unknown, female/heterosexual, female/IDU, and female/other-unknown.

County-level measures

Grounded in the social ecological model as a framework to consider barriers to linkage to HIV care, we assessed county-level community and structural factors representing important social determinants of health including healthcare access, socioeconomic status (SES) and disease burden [12, 13]. County was chosen as the unit of measurement because all of the variables of interest were commonly measured or available in aggregate at this level. The measures were drawn from several sources including a CDC-developed Vulnerability Index (VI) which has helped to identify counties at high risk for incident HIV/HCV cases [14, 15]. The VI is comprised of measures such as percent of the population with a car, below the federal poverty level, who are White, have poor or fair health, are smokers, or have a disability. The VI also assesses per capita income, teen birth rate and HIV prevalence. These measures all represent social determinants that may pose barriers to linkage to care for HIV (Table 1) [15]. We retained all 15 variables from the CDC study and included 63 additional collected from the 2010 US Census, as well as TN state-specific indicators from the CDC and TDH surveillance data (Table 2).

Table 1

Demographics of cohort of HIV-positive patients in Tennessee between 2010–2016.

Demographic Value	Category	Total		30-Day		60-Day		90-Day
		#	%	#	%	#	%	#	%
		(n = 3751)		(n = 1561)		(n = 2266)		(n = 2588)
Sex	Male	2988	80%	1197	77%	1745	77%	2013	78%
	Female	763	20%	364	23%	521	23%	575	22%
Race/ Ethnicity	White (Non-Hispanic)	1231	33%	600	38%	820	36%	935	36%
	Black (Non-Hispanic)	2205	59%	801	51%	1223	54%	1408	55%
	Hispanic (All Races)	200	5%	98	6%	133	6%	150	6%
	Other/ Unknown	115	3%	62	4%	90	4%	95	3%
Age at Diagnosis (years)	Median [IQR]	31	[24,43]	32	[24,45]	32	[24,44]	32	[24,44]
HIV Risk Factor	Heterosexual	883	24%	378	24%	563	25%	635	25%
	MSM	2080	55%	864	55%	1270	56%	1463	57%
	IDU	114	3%	54	3%	74	3%	82	3%
	MSM/IDU	75	2%	37	2%	49	2%	55	2%
	Other/ Unknown	599	16%	228	15%	310	14%	353	14%
Year of Diagnosis	2012	842	22%	333	22%	495	22%	576	22%
	2013	756	20%	333	22%	488	22%	535	21%
	2014	729	19%	314	20%	464	20%	529	20%
	2015	716	19%	305	20%	433	19%	494	19%
	2016	708	19%	276	18%	386	17%	454	18%
Site of Diagnosis	Inpatient Facility or ER	746	20%	397	25%	499	22%	746	20%
	Outpatient Facility	1291	34%	593	38%	799	35%	1291	34%
	Health Department or STD/Family Planning Clinic	1041	28%	349	22%	606	27%	1041	28%
	Blood Bank	134	4%	16	1%	39	2%	134	4%
	Correctional Facility	195	5%	49	3%	88	4%	195	5%
	Other/ Unknown	14	0%	1	0%	4	0%	14	0%
	Missing	330	9%	156	10%	231	10%	330	9%

Table 2

Individual-level factors associated with linkage to HIV care.

	30-day			60-day			90-day
Variable	aRR*	95% Confidence Interval	P-Value	aRR*	95% Confidence Interval	P-Value	aRR*	95% Confidence Interval	P-Value
Year			0.34			0.004			0.03
2012 (ref)	1.00			1.00			1.00
2013	1.09	[0.97, 1.22]		1.07	[0.99, 1.16]		1.02	[0.95, 1.08]
2014	1.05	[0.94, 1.18]		1.04	[0.97, 1.13]		1.02	[0.96, 1.09]
2015	1.08	[0.96, 1.22]		1.03	[0.95, 1.12]		1.01	[0.94, 1.07]
2016	0.98	[0.87, 1.11]		0.91	[0.84, 1.00]		0.92	[0.85, 0.99]
Age at Diagnosis			0.09			0.006			<0.001
20	1.09	[1.01, 1.18]		1.06	[1.00, 1.12]		1.07	[1.02, 1.12]
25	1.01	[0.98, 1.04]		1.00	[0.98, 1.02]		1.00	[0.98, 1.02]
30 (ref)	1.00			1.00			1.00
35	1.03	[0.99, 1.07]		1.05	[1.02, 1.08]		1.05	[1.02, 1.07]
40	1.06	[0.98, 1.15]		1.09	[1.03, 1.15]		1.09	[1.05, 1.14]
45	1.07	[0.98, 1.17]		1.10	[1.04, 1.17]		1.10	[1.05, 1.16]
Race/Ethnicity			<0.001			<0.001			<0.001
White, Non-Hispanic (ref)	1.00			1.00			1.00
Black, Non-Hispanic	0.77	[0.71, 0.83]		0.85	[0.81, 0.90]		0.86	[0.82, 0.90]
Hispanic, All Races	1.06	[0.91, 1.23]		1.03	[0.93, 1.15]		1.01	[0.93, 1.10]
Other/Unknown	1.09	[0.90, 1.31]		1.15	[1.03, 1.29]		1.06	[0.96, 1.16]
Sex/Risk Factor			<0.001			<0.001			<0.001
Male/Heterosexual (ref)	1.00			1.00			1.00
Male/MSM	1.15	[0.99, 1.35]		1.09	[0.98, 1.22]		1.05	[0.96, 1.15]
Male/IDU	1.23	[0.98, 1.55]		1.11	[0.94, 1.30]		1.05	[0.93, 1.20]
Male/Other-Unknown	0.93	[0.77, 1.13]		0.85	[0.74, 0.97]		0.80	[0.71, 0.89]
Female/Heterosexual	1.39	[1.18, 1.65]		1.30	[1.16, 1.46]		1.19	[1.08, 1.30]
Female/IDU	1.04	[0.75, 1.45]		1.05	[0.84, 1.31]		0.92	[0.75, 1.13]
Female/Other-Unknown	1.11	[0.89, 1.39]		1.02	[0.87, 1.19]		1.00	[0.88, 1.13]
Site of Diagnosis			<0.001			<0.001			<0.001
Outpatient (ref)	1.00			1.00			1.00
Inpatient Facility or ER	1.18	1.08, 1.29]		1.09	[1.02,1.17]		1.05	[0.99, 1.11]
Health Department or STD/Family Planning Clinic	0.73	[0.66, 0.81]		0.94	[0.88, 1.01]		0.95	[0.90, 1.01]
Blood Bank	0.28	[0.18, 0.44]		0.49	[0.38, 0.64]		0.53	[0.42, 0.66]
Correctional Facility	0.59	[0.46, 0.76]		0.78	[0.66, 0.91]		0.81	[0.71, 0.92]
Other/Unknown	0.18	[0.03, 1.10]		0.52	[0.23, 1.16]		0.56	[0.28, 1.10]
Missing	1.00	[0.88, 1.13]		1.10	[1.01, 1.20]		1.07	[1.00, 1.14]

*Adjusted for year of diagnosis, sex/exposure category, race/ethnicity and site of diagnosis.

Other county-level variables assessed as measures of healthcare access included: percent of the population without health insurance, rate of mental health (MH) providers, per capita urgent care facilities, and per capita primary care physicians [16-18]. MH providers were collected in the 2010 Census and included psychiatrists, psychologists and licensed clinical social workers specializing in MH care. The rate was calculated as the number of MH providers per 100,000 populations. Measures of community socioeconomic status included: percent of the population unemployed, percent of the population with food security, average number of vacant housing units, average number of female-headed households, and average number of drug-related or violent crimes. Finally, measures of community disease burden included: rates of STI diagnosis, percent of HIV cases due to IDU, and average numbers of poor MH days–a measure of community-level mental distress. Average number of mentally unhealthy days was determined using results from the yearly Behavioral Risk Factor Surveillance System (BRFSS) survey that asks participants “… thinking about your mental health, … how many days during the past 30 days was your mental health not good?” [19]. The BRFSS averages the response to this question at the county-level in accordance with its stratified, probabilistic sampling scheme.

Individuals were assigned exposure status to county-level factors based on county of residence at the time of diagnosis. Additionally, counties with fewer than five PLWH were suppressed by the TDH according to data privacy regulations. We hypothesized that county-level measures of healthcare access (percentage without healthcare insurance, rate of MH providers, per capita urgent care facilities, and per capita primary care physician, percentage of homes with cars), SES (percentage below the federal poverty level, percentage unemployed, average vacant housing units, average number of female-headed households, percentage with food insecurity, violent crime rate), and disease burden (percentage of adults with poor or fair health, percentage of adult smokers, percentage with disability, HIV prevalence, rate of STD diagnosis) would be associated with linkage to HIV care.

Data analysis

Individual-level analysis

Descriptive statistics for demographic and clinical characteristics (median, interquartile range [IQR] or percent, as appropriate) were calculated by linkage to care status within 30, 60 and 90 days of HIV diagnosis, and overall. We used modified Poisson regression to assess risk ratios (RR) for linkage to care at each threshold (30, 60, and 90-days) adjusting for a priori selected individual-level covariates in multivariable analysis that are known to be associated with the outcome of interest, including year of- and age at diagnosis, sex, race/ethnicity and HIV risk factor. In the primary analyses, year was modeled as a categorical variable. Age was modeled as a continuous variable using restricted cubic splines with 4 knots to avoid linearity assumptions. Sensitivity analysis was done in which year was modeled as a continuous variable.

County-level analysis

Individuals were assigned to county-level factors based on county of residence at the time of diagnosis (by merging individual eHARS and county-level data). Modified Poisson regression was used to obtain adjusted RR and marginal probabilities with 95% confidence intervals for the association between county-level characteristics and individual-level linkage outcomes. The models were fit at the individual level, incorporating county-level factors by treating individuals as being nested within counties (and therefore uniformly exposed within counties). We adjusted for individual-level age, sex, race/ethnicity, HIV transmission risk, site of diagnosis and time since HIV diagnosis. These covariates were modeled using restricted cubic splines for continuous measures and categorical indicators for all other measures. County-level factors were included in multivariable models based on a priori identification from CDC’s vulnerability index factors, factors associated with healthcare access, and socioeconomic factors as descrived above. We conducted pairwise correlation of all county-level variables and among those that were highly correlated (e.g., correlation < -0.8 or >0.8) only one factor was included to avoid collinearity. Robust standard errors for all models were calculated by clustering at the county level, assuming correlation in the primary outcomes between individuals residing in the same county at the time of HIV diagnosis.

Results

Description of cohort of TN residents newly diagnosed PLWH

The data included 3,751 newly diagnosed PLWH in TN between 2012 and 2016. The number of newly diagnosed PLWH gradually decreased from 2012 to 2016 (2012: 842, 2013: 756, 2014: 729, 2015: 716, 2016: 708). Men comprised a greater proportion of the cohort (80%, n = 2988) than women; and Black patients (59%, n = 2205) comprised a greater proportion of the cohort than White (33%, n = 1231) or Hispanic patients (5%, n = 200). The median age at diagnosis was 31 years [IQR 24, 43]. Over half (55%, n = 2080) of the population reported a transmission risk factor of MSM, while 24% (n = 883) reported heterosexual sex and 3% (n = 114) reported IDU. More patients were diagnosed from outpatient facilities (34%, n = 1291) than health department or STD clinics (28%, n = 1041), inpatient facilities or ERs (20%, n = 746), correctional facilities (5%, n = 195) or blood banks (4%, n = 134) (Table 1). Four counties in TN represented 71% of incident cases during the analysis period (Shelby County, county seat of Memphis (n = 1460, 39%); Davidson County, county seat of Nashville (n = 784, 21%); Hamilton County, county seat of Chattanooga (n = 232, 6%); and Knox County, county seat of Knoxville (n = 201, 5%)).

Trends in establishing HIV care over time

Over the study period, 42% (n = 1,561) of newly diagnosed PLWH were linked to care within 30-, 60% (n = 2,266) within 60-, and 69% (n = 2,588) within 90-days. The proportion of patients linked to care within 30 days of diagnosis increased from 40% (n = 333) in 2012 to 44% (n = 333) in 2013 and decreased to 40% (n = 276) in 2016. Whether linkage to HIV care was defined at 30, 60, or 90 days after HIV diagnosis, linkage increased from 2012 to 2013 then declined to or below the 2012 value by the end of the study period in 2016. As the time to linkage threshold was broadened, the percentage of patients who were linked to an HIV provider increased. Adjusting for other patient-level factors (age, sex, transmission risk factor, site of diagnosis), 30-day linkage to care increased by 13% (aRR 1.13, 95%CI 1.03–1.24), and 60-day linkage to care increased by 9% (aRR 1.09, 95%CI 1.02–1.16) in 2013 compared to 2012. However, the adjusted rate of 30-, 60- and 90-day linkage to care did not significantly differ in 2014, 2015, or 2016 compared to 2012 (with the exception of the risk of 90-day linkage to care in 2016, which decreased by 7% compared to 2012 (aRR 0.93, 95% CI 0.87–0.99)) (Fig 1). When modeled as a linear covariate, year of diagnosis was not a significant predictor of 30-day linkage to care (aRR 0.99, 95% CI 0.97–1.01).

Fig 1

The rates of 30, 60, and 90-day linkage to HIV Care in Tennessee for patient diagnoses between 2012 and 2016.

Individual-level predictors of linkage to care

Age was a good individual predictor of linkage to care. Younger and older patients were more likely to establish care within 30 days (compared to 30 year-olds, aRR 1.09, 1.01, 1.03, 1.06, and 1.07, respectively for ages 20, 25, 35, 40, and 45 years). Race was also an independent predictor of linkage to care. Black patients had a significantly decreased rate of 30-day linkage to care compared to Whites (aRR 0.77, 95% CI 0.71–0.83). Linkage to care did not differ significantly between White and Hispanic (aRR 1.06, 95%CI 0.91–1.23) or other/unknown patients (aRR 1.09, 95%CI 0.90–1.31). When we combined sex and HIV transmission risk factor categories, we found that heterosexual females (aRR 1.39, 95%CI 1.18–1.65) were more likely to link to HIV care than heterosexual males. Additionally, the location of HIV testing/diagnosis was an important predictor of linkage to care. Compared to an inpatient facility or emergency room, patients diagnosed at inpatient facilities (aRR 1.18, 95%CI 1.08–1.29) were more likely and patients diagnosed at health departments or STD clinics (aRR 0.73, 95%CI 0.66–0.81) and correctional facilities (aRR 0.59, 95%CI 0.46–0.76) were less likely to establish HIV care (Table 2). While the data here are presented for 30-day linkage to care, the same patterns were seen for 60- and 90-day linkage to care as illustrated in Table 2.

County-level factors associated with linkage to care

Pair wise comparison of association between county-level variables revealed a substantial amount of collinearity. Among the 29 county-level measures assessed, 12 were highly correlated and not included in the model. Accordingly, 17 measures remained in the multivariable model. Only two variables were both clinically and statistically significant in multivariable analysis: Average poor mental health was the strongest county-level predictor of poor linkage care at 30 days (aRR 0.63, 95%CI: 0.40–0.99 per 10-unit increase in poor mental health days). Teen birth rate was also significantly associated with individual linkage to care at 30 days (aRR 1.02, 95%CI: 1.01, 1.04 per 10% increase). For every 10% increase increase in HIV cases due to IDU, individual linkage to care decreased by 4% (aRR 0.91, 95%CI: 0.91–1.00), but this variable did not meet the threshold for statistical significance (Table 3). If one does a Bonferoni adjustment for multiple comparisons, none of the county-level factors remains statistically significant. Notably, White/Non-White segregation index, a variable that reflects greater residental segregation between non-White and White county residents was not included in the final model, but was highly correlated with five of the variables included in final model. Also, in this model which adjusted for both individual and county level factors, White and Hispanic individuals had an increased risk of 30-day linkage to care compared to Black individuals (aRR 1.33, 95%CI 1.30–1.37, aRR 1.44, 95% CI 1.41–1.47 respectively) [data not shown].

Table 3

County level predictors of linking to HIV care within 30 days of diagnosis in Tennessee.

Factor	RR [95% CI]
Avg. Monthly mental unhealthy days (per 10)	0.63* [0.40–0.99]
Avg. Morphine milligram equivalent (per 1000)	0.99 [0.98–1.01]
Avg. no. drug-related crimes (per 100)	1.00 [0.99–1.01]
Avg. no. drug-related deaths (per 10)	1.01 [0.94–1.08]
Drug trafficking hot-zone	3.37 [0.88–12.89]
No. methadone clinics	1.06 [0.93–1.20]
Per capita income (log10)	2.95 [0.35–24.78]
Per capita primary care physicians (per 10%)	0.95 [0.74–1.22]
Per capita urgent care facilities (per 10%)	0.51 [0.10–2.60]
Percent below FPL	1.47 [0.07–29.05]
Percent of adults smoking (per 10%)	0.99 [0.86–1.13]
Percent of HIV cases due to IDU (per 10%)	0.96 [0.91–1.00]
Percent unemployed	1.01 [0.96–1.06]
Percent with poor/fair health	1.16 [0.98–1.38]
Percent without health insurance	0.98 [0.94–1.03]
Rate mental health providers (per 10%)	1.05 [0.93–1.20]
Teen birth rate (per 10%)	1.02* [1.01–1.04]
* p<0.05

*Risk Ratio adjusted for age, sex, race/ethnicity, transmission risk factor, and site of diagnosis.

Avg = Average; No = Number; FPL = Federal poverty line.

IDU = Intravenous drug use; STD = Sexually Transmitted Diseases.

Linkage to care in the highest burden counties in TN

We analyzed the marginal probabilities of linkage at 30-days in the four highest-burden metropolitan counties by race/ethnicity and found that Black patients persistently had the lowest probability of 30-day linkage to care as compared to both White and Hispanic individuals when adjusting for individual level factors, and when adjusting for both individual and county-level factors and when interacting individual-level race/ethnicity with county of residence (Fig 2). Racial disparities were least prominent in Davidson County (the county seat of Nashville), whose residents also had the highest probability of linkage to care of the four highest-HIV-burdened counties in TN.

Fig 2

The probability of linking to HIV care within 30 days of diagnosis by race/ethnicity for patients living in the four counties with the highest burden of HIV in Tennessee.

As in the entire cohort, more patients were diagnosed in outpatient facilities (n = 1025, 38%) than other sites. Thirty-day linkage to care from outpatient facilities was poor across all of 4 counties and ranged from 44% to 53%, and 30-day linkage to HIV care from inpatient or ER facilities ranged from 43% to 59% in each of the four high-burden metropolitan counties showed that the proportion of patients linked to care was more variable and ranged from 43% to 59% (Table 4).

Table 4

County level 30-day linkage to HIV care rates by county and facility type in Tennessee.

Facility Type of HIV Diagnosis	Combined		Shelby		Knox		Hamilton		Davidson
	N	%	N	%	N	%	N	%	N	%
Inpatient Facility or Emergency Room	479	48.85%	276	43.48%	33	57.58%	34	58.82%	135	55.14%
Outpatient Facility	1025	45.46%	591	44.16%	53	52.83%	102	50.00%	279	45.16%
Health Department or STD/Family Planning Clinic	713	33.94%	315	38.10%	83	34.94%	57	45.61%	258	25.97%
Blood Bank	110	10.90%	84	9.52%	4	0.00%	8	37.50%	14	7.14%
Correctional Facility	169	21.30%	109	18.35%	24	25.00%	1	0.00%	35	28.57%
Other/Unknown	10	10.00%	6	0.00%	1	100.00%	0	-	3	00.00%
Missing	171	47.37%	79	44.30%	3	66.67%	30	53.33%	59	47.46%

Discussion

Our analysis of patterns and predictors of linkage to HIV care in TN between 2012 and 2016 highlights unsettling trends. First, despite concerted efforts from TDH, CDC and local partners, timely linkage to HIV care among newly diagnosed PLWH in TN has not only failed to improve over time, but TN now trails the nation in linking PLWH to care. Second, unacceptable racial disparities in linkage to care persist, as Blacks remain much less likely to link to care than Whites–even after accounting for a wide range of individual and structural factors that often are drivers of poor healthcare access and engagement. Our analysis has contributed to the growing call to operationalize measures of structural racism impacting health outcomes by identifying some potential systematic and programmatic opportunities that could be areas for intervention to begin to change this trend in TN [20]. At the same time, our analysis highlights the difficulty of measuring this factor. Indeed, while we assessed 29 county-level variables that all represent social determinants of health, few were statistically significant predictors of individual linkage to care.

In addition to the importance of individual factors like race/ethnicity, our analysis also underscores site of diagnosis as a key predictor of linkage to HIV care. Most studies have highlighted lower linkage to care at testing sites without co-located medical facilities despite higher positivity rates in non-healthcare settings [21-23]. Individuals in our cohort diagnosed at sites without co-located medical facilities, such as correctional facilities and blood banks, were the least likely to link to HIV care. While outpatient facilities yielded the greatest numbers of incident diagnoses in this cohort, inpatients facilities had an 18% increased likelihood of linkage to care. Higher likelihood of linkage to HIV care from inpatient facilities may also reflect the fact that patients diagnosed in these settings are more ill, and thus will more readily establish care after hospitalization. Alternatively, this finding could be artifactual–reflecting routine disease staging with CD4 count and viral load aseessment after diagnosis in the inpatient setting, and not in fact, linkage to care. Nonetheless, linkage to care from both inpatient facilities and outpatient facilities was lowest in Shelby County, the County seat of Memphis, and TN’s only priority county nationally targeted for EtE activities. Such findings represent an opportunity for improvement via optimization of linkage referrals and implementation of models such as rapid treatment initiation to promote earlier linkage to care [24].

Our analysis of county-level drivers of linkage to HIV also yielded some intriguing findings. The strongest county-level predictor of linkage to HIV care in TN was the average monthly number of mentally unhealthy days. While a rich body of data supports an association between community factors and health outcomes, much of this research has focused on poor SES and other characteristics of neighborhood deprivation [13, 25, 26]. One study set in TN found that individuals living in neighborhoods with the most adverse SES were least likely to achieve virologic suppression [13]. Some authors hypothesize that the relationship between SES and health outcomes are mediated by distribution of stressors, which may be more prevalent in poorer neighborhoods and among racial and ethnic minorities [27, 28]. Others suggest that maladaptive response to stressors may disproportionately impact those with low SES [28]. In our analysis, this association between county mental health and linkage to care was strong, and independent of race.

Surprisingly, access to health insurance at the county level was not significantly associated with linkage to care. While TN has not expanded Medicaid, through the federally-supported Ryan White (RW) program, TN is still able to provide coverage for medical services associated with HIV/AIDS and related illnesses, general insurance assistance and treatment coverage [29]. Across the country, recipients of these funds are more likely to succeed along the continuum of care when compared to uninsured PLWH or those with other forms of healthcare coverage [30, 31]. As such, the promotion of RW services in TN may be an important mechanism to address unmet mental health needs as earlier described. Interestingly, increases in teen birth rate were associated with a small, statistically significant increase in linkage to HIV care. The reasons for this are not entirely clear, but could reflect intense wrap around services for pregnant and peripartum women to prevent mother-to-child HIV transmission or could reflect potential confounding.

It is well-documented that HIV disproportionately affects the Black community in the US at large–a disparity that has persisted decades since the start of the HIV epidemic [6, 8, 9]. Our study findings add to the literature highlighting a critical need to adopt comprehensive strategies to measure drivers of persistent and pervasive racial disparities in HIV outcomes to guide improvement. Today, the life-changing pandemic caused by the novel SARS-coronavirus-2 has targeted a floodlight on the power of structural racism to undermine public health as whole, and to precipitate disparities in COVID-19 infection, hospitalization and death [32, 33]. These trends have furthered important discussions about systemic racial disparities in the US healthcare system and may afford a critical opportunity to seriously consider how to address the structural factors driving such disparities in our healthcare system. In our analysis, the fact that racial disparities persisted despite accounting for both individual characteristics, and as aggregated at the county-level among the four highest-burdened metropolitan areas, speaks to the insidious and complex nature of structural racism. Additionally, the high correlation of residential racial segregation (White, non-White) with many county level factors, while not surprising, further underscores the relationship between race and a range of geographic factors that can impact health. Some, like former president of the American Public Health Association, Dr. Camara Jones, have called on us to recognize structural racism as “a system of structuring opportunity and assigning value based on the social interpretation of how one looks;” and the root cause of all differences in any health outcome associated with race [34]. As such, racism is an important social determinant of health that necessitates a structural intervention [34]. Acknowledging these complex dynamics, several American cities have declared racism as a public health crisis and committed to put racial equity at the core of all city procedures to advocate for policies that improve health in communities of color [35, 36]. Other cities and counties have made similar declarations, but it is clear that they must be accompanied by novel structural approaches to effectively reduce these disparities, and ultimately end the HIV epidemic [37].

Our analysis has notable strengths and weaknesses. Our integration of individual-level surveillance with county-level census and publicly reported data allowed us to identify important individual and county-level risk factors for poor linkage to care, while accounting for many of the socioeconomic drivers of racial disparities in health. However, our use of such data also posed some limitations, as we could not incorporate important factors not readily collected in these systems like individual mental health, experiences with stigma, racism, SES and other barriers to healthcare access and earlier linkage to HIV care for PLWH in TN. We were also unable to distinguish transgender individuals who are at great risk for poor health outcomes. Additionally, despite the improvements in HIV surveillance and data quality since 2012, our measures of linkage to care were reliant on the completeness of the mandatory reporting system, which varies by site and could have introduced some bias. We acknowledge that we have included many covariates in our analyses, and could be subject to limitations from multiple testing. Finally, more granular spatial analysis (i.e. ZIP Code rather than county-level) was limited by both data suppression requirements for TDH and a lack of available public health data at the zip-code level.

Conclusions

In conclusion, to meet the critical EtE target of reducing the number of new HIV infections in the US by 90% and move towards ending the epidemic, statewide linkage to care in TN needs to improve. Despite targeted efforts both broadly and in minority communities, linkage to HIV care did not improve substantially from 2012 to 2016. Racial disparities that persist at both individual and county levels suggest the need for exploring structural interventions to address racism as a public health threat. In addition, optimizing outreach for young heterosexual men who may be overlooked by interventions targeting MSM, and addressing linkage to care processes from outpatient and community-based testing facilities through improved partnerships or co-location of testing and treatment services are potential areas for intervention. Further exploration of the role of poor community and individual mental health in this environment is needed to inform mental health interventions to improve engagement in HIV care.

22 Jul 2021

PONE-D-21-13112

Individual, Community, and Structural Factors Associated with Linkage to HIV Care Among People Diagnosed with HIV in Tennessee

PLOS ONE

Dear Dr. Ahonkhai,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. In your response, please pay particular attention to the concerns regarding the definition of linkage being used in the paper, as well as clarification of the modeling approach and interpretation of the study findings in light of structural factors.

Please submit your revised manuscript by Sep 03 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see:  http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols . Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at  https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols .

We look forward to receiving your revised manuscript.

Kind regards,

Becky L. Genberg

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. In your Methods section, please ensure that the data sources and codes used are described in adequate detail.

3. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see http://journals.plos.org/plosone/s/data-availability.

Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories. Any potentially identifying patient information must be fully anonymized.

Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access.

We will update your Data Availability statement to reflect the information you provide in your cover letter.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Reviewer #1: Overview:

This study evaluates individual and county-level predictors of linkage to HIV care following new HIV diagnosis in Tennessee over time. The manuscript addresses an important topic and adds particular value by identifying structural drivers of poor linkage and persistent racial disparities using county-level data. The paper is well written and I think makes an important contribution, though I have some concerns about the definition of linkage to care and its interpretation. My comments/suggested edits are below:

Major:

1. The authors define linkage as the date that the first CD4 count or HIV viral load measurement is taken including those drawn on the same date that the HIV diagnosis was made. While this certainly could represent linkage to HIV care, particularly if the diagnosis was made in an outpatient facility that also provides chronic HIV care, it equally could represent labs that are drawn in a non-HIV care facility (e.g. emergency department or inpatient hospital) at the time the diagnosis was made. Given that this likely does not constitute true linkage, I think the authors should consider incorporating sensitivity analyses where CD4 count/viral load measures within 24 hours of diagnosis from inpatient/emergency department or other episodic/urgent care settings (e.g. blood banks) are excluded from the linkage definition. Additional more granular information about the time to linkage following diagnosis would also be helpful – how commonly was the definition of linkage met on the same day as the diagnosis and how did this differ by care setting?

2. Given the above uncertainties with the linkage definition, I suggest incorporating a second outcome of the proportion achieving viral suppression by 6 and/or 12 months after diagnosis. This measures a different process than linkage alone, however it would strengthen study findings, particularly if disparities are similar between those who are less likely to link and those less likely to attain viral suppression.

Minor:

1. Line 46 – consider changing ‘performing’ to something like ‘experience worse outcomes’. The wording here implies that non-Hispanic Black individuals are to blame for ‘underperforming’ in the care cascade, whereas the authors clearly intend to shed light on racial disparities and ways that the healthcare system is underperforming for these individuals.

2. Line 54 – “US” is duplicated, remove one instance.

3. Some of the county-level factors appear to have very extreme point estimates (e.g. ‘percent experiencing food insecurity’ with RR of 4.7 million and ‘percent households with a car with RR of 0.00). Can the authors double check these results and provide some explanation for the extreme values here?

Reviewer #2: The authors assessed trends and factors associated with linkage to care among TN residents diagnosed with HIV from 2012 to 2016. The article was well-written. And the incorporation of a vast number of community-level exposures is a nice addition to the literature. My comments are detailed below.

Abstract. The methods and discussion in the abstract suggest you ran county-level models. However, based on the methods in the narrative, it seems you ran a county-level analysis using individual data (i.e., using individual data with county-level independent variables). Is this an ecologic analysis?

Methods: In general, I think the modelling approach requires clarification. And the approach for incorporating the county-level variables requires clarification.

1. Study, setting and design. How were county-level measures assigned or merged with eHARS data? Is the variable merged to an individual based on county of diagnosis or county of residence? Is this ecologic data: do you have measure county-level linkage and merge with a county-level exposure dataset?

2. Measures. For a variable like poor mental health days, did you use the average for the county? How variable is the measure – would the median be a better metric for a county?

3. Measures. Are all of the community SES variables a percent in the county? And were you able to drill down further, say to the zipcode?

4. Individual-level analysis. What are the a priori covariates in the multivariable analysis? And how/why were they chosen? Based on the tables, I think you fit three models, one for each threshold, with the variables listed under table 2, but this could be clearer in the narrative.

5. County-level analysis. You have listed a lot of independent variables (N = 23 county level). Did you fit a model for each independent variable/exposure of interest? If not, how did you address collinearity? Current approach suggests a need to correct for a lot of multiple testing – how was this addressed?

6. County-level analysis. How were model covariates selected? And was the approach for selecting model covariates different than the approach for selecting model covariates in the individual-level analysis. If so, why?

Discussion. My primary concern with this discussion is that you’ve zeroed in on the significant findings and given very little consideration to your mostly null county-level associations. Usually, I think that’s fine, but in this case, I worry a reader may consider it fishing for a county-level association. Is it possible that the county-level variables are measured at too wide of a geographic level, e.g., would a zip-code level exposure be better?

1. The inpatient finding seems artifactual, given persons diagnosed while at an inpatient facility are admitted / on-site. I might re-frame this to look at outpatient as your reference. It seems more actionable to know if the health department or correctional facilities perform as well as outpatient facilities (table 2 suggests they perform worse).

2. Why declining linkage by 2016? What was happening or stopped happening – any TN DOH initiatives?

3. Minor / editorial, but I would suggest using a word other than “incited” to highlight the increase in discussions about racial disparities (discussion, paragraph 2). Is it possible to cite HIV literature on structural determinants versus COVID?

4. The structural racism angle/paragraph requires additional work. Your paper is about individual and community exposures. How do you tie them to structural racism (as a root cause)? And largely, your community level measures had a null association with linkage. Does this support your theory of structural racism as a root cause? Do you consider these county measures proxies for structural racism?

5. Limitations. I would add multiple testing. Are county level variables too broad of an exposure?

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

29 Nov 2021

Editor Comments

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming.

We have edited the manuscript to ensure that it meets PLOS ONE’s style requirements.

2. In your Methods section, please ensure that the data sources and codes used are described in adequate detail.

We have updated the methods section to ensure that the data sources and codes are clearly described.

3. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see http://journals.plos.org/plosone/s/data-availability. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories. Any potentially identifying patient information must be fully anonymized.

These data cannot be shared publicly according to CDC HIV/STD policy and Tennessee law.

4. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter.

CDC HIV/STD security and confidentiality policy and Tennessee law prohibits the public release of de-identified, client-level HIV surveillance data. Publicly available information on persons newly diagnosed with HIV can be found here: https://www.tn.gov/health/health-program-areas/statistics/health-data/hiv-data.html. We appreciate you updating this information in our Data Availability statement.

Reviewer 1 Comments

1. The authors define linkage as the date that the first CD4 count or HIV viral load measurement is taken including those drawn on the same date that the HIV diagnosis was made. While this certainly could represent linkage to HIV care, particularly if the diagnosis was made in an outpatient facility that also provides chronic HIV care, it equally could represent labs that are drawn in a non-HIV care facility (e.g. emergency department or inpatient hospital) at the time the diagnosis was made. Given that this likely does not constitute true linkage, I think the authors should consider incorporating sensitivity analyses where CD4 count/viral load measures within 24 hours of diagnosis from inpatient/emergency department or other episodic/urgent care settings (e.g. blood banks) are excluded from the linkage definition. Additional more granular information about the time to linkage following diagnosis would also be helpful – how commonly was the definition of linkage met on the same day as the diagnosis and how did this differ by care setting?

We agree with the challenges posed with our initial diagnosis of linkage, and have updated the definition. Linkage is now defined at the date of the first CD4 count or viral load after HIV diagnosis. The methods have been updated (Page 3, Lines 72-74) to reflect this. The results have not substantially changed (neither effect size nor direction of responses).

The outcome of interest was linkage to HIV care, which was defined as receipt of the first CD4 or HIV-1 RNA test result captured via TN’s enhanced HIV/AIDS reporting system (eHARS) after diagnosis, and was assessed at 30, 60, and 90 days.

2. Given the above uncertainties with the linkage definition, I suggest incorporating a second outcome of the proportion achieving viral suppression by 6 and/or 12 months after diagnosis. This measures a different process than linkage alone, however it would strengthen study findings, particularly if disparities are similar between those who are less likely to link and those less likely to attain viral suppression.

As described in the response to Reviewer 1 Comment 1, we have updated the definition of linkage to care. Linkage is now defined at the date of the first CD4 count or viral load after HIV diagnosis. The methods have been updated (Page 3, Lines 72-74) to reflect this. The results have not substantially changed (neither effect size nor direction of responses).

3. Line 46 – consider changing ‘performing’ to something like ‘experience worse outcomes’. The wording here implies that non-Hispanic Black individuals are to blame for ‘underperforming’ in the care cascade, whereas the authors clearly intend to shed light on racial disparities and ways that the healthcare system is underperforming for these individuals.

We have edited this text as suggested (Page 2, Line 47) which now reads as below:

In addition, racial/ethnic disparities have persisted across the US over decades, with non-Hispanic Black (Black) individuals typically experience worse outcomes than other racial/ethnic groups across the entire continuum of HIV care.

4. Line 54 – “US” is duplicated, remove one instance.

We have fixed this typographical error.

5. Some of the county-level factors appear to have very extreme point estimates (e.g. ‘percent experiencing food insecurity’ with RR of 4.7 million and ‘percent households with a car with RR of 0.00). Can the authors double check these results and provide some explanation for the extreme values here?

These extreme results were due to collinearity in our analyses (see response to Reviewer 2, comment 6). We have updated the analysis to account for collinearity among county-level factors, by excluding some county level factors that are highly correlated with each other. Withthis approach, estimates are more stable. Also, percent households with a car is no longer included in the model.

Reviewer 2 Comments:

1. Abstract. The methods and discussion in the abstract suggest you ran county-level models. However, based on the methods in the narrative, it seems you ran a county-level analysis using individual data (i.e., using individual data with county-level independent variables). Is this an ecologic analysis?

We appreciate the opportunity to clarify this point. This is not an ecological analysis, as we assessed individual-level outcomes. We have updated the methods section in the abstract (Page 1, Lines 22-25) to clarify the modeling/analysis approach.

TN residents diagnosed with HIV from 2012-2016 were included in the analysis (n=3,751). Individuals were assigned to county-level factors based on county of residence at the time of diagnosis. Linkage was defined by the first CD4 or HIV RNA test date after HIV diagnosis. We used modified Poisson regression to estimate probability of 30-day linkage to care at the individual-level, and the contribution of individual and county-level factors to this outcome.

2. Methods: In general, I think the modelling approach requires clarification. And the approach for incorporating the county-level variables requires clarification. Study, setting and design. How were county-level measures assigned or merged with eHARS data? Is the variable merged to an individual based on county of diagnosis or county of residence? Is this ecologic data: do you have measure county-level linkage and merge with a county-level exposure dataset?

We have updated the methods (Page 5, Lines 144-158) to clarify the modeling approach, and how individual and county-level measures were merged. This section now reads as below:

Individuals were assigned to county-level factors based on county of residence at the time of diagnosis (by merging individual eHARS and county-level data). Modified Poisson regression was used to obtain adjusted RR and marginal probabilities with 95% confidence intervals for the association between county-level characteristics and individual-level linkage outcomes. The models were fit at the individual level, incorporating county-level factors by treating individuals as being nested within counties (and therefore uniformly exposed within counties). We adjusted for individual-level age, sex, race/ethnicity, HIV transmission risk, site of diagnosis and time since HIV diagnosis. These covariates were modeled using restricted cubic splines for continuous measures and categorical indicators for all other measures. County-level factors were included in multivariable models based on a priori identification from CDC’s vulnerability index factors, factors associated with healthcare access, and socioeconomic factors as descrived above. Models were fit including all county-level variables from each of these domains separately, and then together across all county-level domains in a full multivariable model. Robust standard errors for all models were calculated by clustering at the county level, assuming correlation in the primary outcomes between individuals residing in the same county at the time of HIV diagnosis.

3. Measures. For a variable like poor mental health days, did you use the average for the county? How variable is the measure – would the median be a better metric for a county?

The Behavioral Risk Factor Surveillance Survey provides average results at the county level (as well as results averaged across other demographic factors). It is not possible to extract different measures of central tendency at this stage, as this county-level averaging is done centrally after applying the appropriate weights to each surveillance sample. We have updated the methods section (Page 4, Lines 116-118) to reflect this as below:

Average number of mentally unhealthy days was determined using results from the yearly Behavioral Risk Factor Surveillance System (BRFSS) survey that asks participants “… thinking about your mental health, … how many days during the past 30 days was your mental health not good?”19 The BRFSS averages the response to this question at the county-level, in accordance with its stratified, probabilistic sampling scheme.

4. Measures. Are all of the community SES variables a percent in the county? And were you able to drill down further, say to the zipcode?

County was chosen as the unit of geographic analysis for two reasons. First, some of the geographic measures were available at the county but not ZIP-Code-level. Second, due to data privacy policies, Tennessee Department of Health could not share patient-level data for regions with fewer than 5 HIV cases, hence using a smaller geographic area would have created more missing data in the analysis. We have updated the methods (Pages 3 and 4, Lines 93-95 and 120-122), and descrbied this as a limitation in the Discussion (Page 15, Lines 345-347).

Methods

Grounded in the social ecological model as a framework to consider barriers to linkage to HIV care, we assessed county-level community and structural factors representing healthcare access, socioeconomic status (SES) and disease burden.12,13 County was chosen as the unit of measurement because all variables of interest were commonly measured or available in aggregate at this level. The measures were drawn from several sources including a CDC-developed Vulnerability Index (VI) which has helped to identify counties at high risk for incident HIV/HCV cases.14,15 The VI is comprised of measures such as percent of the population with a car, below the federal poverty level, who are White, have poor or fair health, are smokers, or have a disability. The VI also assesses per capita income, teen birth rate and HIV prevalence. These measures all represent social determinants that may pose barriers to linkage to care for HIV [Table 1].15 We retained all 15 variables from the CDC study and included 63 additional collected from the 2010 US Census, as well as TN state-specific indicators from the CDC and TDH surveillance data [Table 2].

Individuals were assigned exposure status to county-level factors based on county of residence. Additionally, counties with fewer than five PLWH were suppressed by the TDH according to data privacy regulations.

Discussion

Finally, more granular spatial analysis (i.e., ZIP Code rather than county-level) was limited by both data suppression requirements for TDH, and lack of available public health data at the zip-code level.

5. Individual-level analysis. What are the a priori covariates in the multivariable analysis? And how/why were they chosen? Based on the tables, I think you fit three models, one for each threshold, with the variables listed under table 2, but this could be clearer in the narrative.

Individual-level covariates were limited to available surveillance data collected. These demographic factors (age, sex, race-ethnicity, transmission risk factor) are also often associated with HIV care outcomes, so all were included in the analysis. We have updated the methods section (Page 5, Lines 135-141) to better explain the analysis plan.

We used modified Poisson regression to assess risk ratios (RR) for linkage to care at each threshold (30, 60, and 90-days) adjusting for a priori selected individual-level covariates in multivariable analysis that were available in the surveillance data and known to be associated with the outcome of interest, including year of and age at diagnosis, sex, race/ethnicity and HIV transmission risk factor.

6. County-level analysis. You have listed a lot of independent variables (N = 23 county level). Did you fit a model for each independent variable/exposure of interest? If not, how did you address collinearity?

Our previous analyses did not account for collinearity, which was an oversight by us and has been fixed in the revised manuscript. Many of these county-level variables are highly correlated. In our revised analysis, we looked at the pairwise correlation of all county-level variables and among those that were highly correlated (e.g., correlation < -0.8 or >0.8) we selected one county-level variable to include in the model; our choice was based on our perceived scientific interest. The Methods (Page 5, Lines 154-158) and Results (Page 11, Lines 217-229) sections have been updated accordingly.

Methods

We conducted pairwise correlation of all county-level variables and among those that were highly correlated (e.g., correlation < -0.8 or >0.8) only one factor was included to avoid collinearity.

Results

Pair wise comparison of association between county-level variables revealed a substantial amount of collinearity. Among the 29 county-level measures assessed, 12 were highly correlated and not included in the model. Accordingly, 17 measures remained in the multivariable model. Only two variables were both clinically and statistically significant in multivariable analysis: Average poor mental health was the strongest county-level predictor of poor linkage care at 30 days (aRR 0.63, 95%CI: 0.40-0.99 per 10-unit increase in poor mental health days). Teen birth rate was also significantly associated with individual linkage to care at 30 days (aRR 1.02, 95%CI: 1.01, 1.04 per 10% increase). For every 10% increase increase in HIV cases due to IDU, individual linkage to care decreased by 4% (aRR 0.91, 95%CI: 0.91-1.00), but this variable did not meet the threshold for statistical significance [Table 3]. If one does a Bonferoni adjustment for multiple comparisons, none of the county-level factors remains statistically significant. Notably, White/Non-White segregation index, a variable that reflects greater residental segregation between non-White and White county residents was not included in the final model, but was highly correlated with five of the variables included in final model [data not included].

Table 3: County level predictors of linking to HIV care within 30 days of diagnosis in Tennessee

Factor RR [95% CI]

Avg. Monthly mental unhealthy days (per 10) 0.63* [0.40-0.99]

Avg. Morphine milligram equivalent (per 1000) 0.99 [0.98-1.01]

Avg. no. drug-related crimes (per 100) 1.00 [0.99-1.01]

Avg. no. drug-related deaths (per 10) 1.01 [0.94-1.08]

Drug trafficking hot-zone 3.37 [0.88-12.89]

No. methadone clinics 1.06 [0.93-1.20]

Per capita income (log10) 2.95 [0.35-24.78]

Per capita primary care physicians (per 10%) 0.95 [0.74-1.22]

Per capita urgent care facilities (per 10%) 0.51 [0.10-2.60]

Percent below FPL 1.47 [0.07-29.05]

Percent of adults smoking (per 10%) 0.99 [0.86-1.13]

Percent of HIV cases due to IDU (per 10%) 0.96 [0.91-1.00]

Percent unemployed 1.01 [0.96-1.06]

Percent with poor/fair health 1.16 [0.98-1.38]

Percent without health insurance 0.98 [0.94-1.03]

Rate mental health providers (per 10%) 1.05 [0.93-1.20]

Teen birth rate (per 10%) 1.02* [1.01-1.04]

* p<0.05

7. Current approach suggests a need to correct for a lot of multiple testing – how was this addressed?

We acknowledge that we have included many covariates in our analyses, and that there may be an issue with multiple testing. In general, we usually do not like to adjust for multiple comparisions because it is not straightforward to determine exactly how many comparisons were performed (does one count per outcome [30-day, 60-day, or 90-day linkage], or per table, or over the entire manuscript). Our preference is to present all results and then to let reviewers decide how to interpret them. However, we agree that it is worth pointing out the multiple comparisons to readers. In the revision, we added a sentence in the Results when discussing the county-level factors (Page 11, Lines 225-227):

If one does a Bonferoni adjustment for multiple comparisons, none of the factors remains statistically significant.

8. County-level analysis. How were model covariates selected? And was the approach for selecting model covariates different than the approach for selecting model covariates in the individual-level analysis. If so, why?

County-level variables were purposefully selected to reflect healthcare access, community socioeconomic status, and community disease burden. We also included all 15 variables in the CDC-developed Vulnerability Index (VI) which has helped to identify counties at high risk for incident HIV/HCV cases. We hypothesized that these variables would be associated not only with new HIV infection, but also with linkage toHIV care. All variables included reflect important social determinants of health. We have edited the methods (Page 3, Lines 91-93) to clarify this.

Grounded in the social ecological model as a framework to consider barriers to linkage to HIV care, we assessed county-level community and structural factors representing important social determinants of health including healthcare access, socioeconomic status (SES) and disease burden.

9. Discussion. My primary concern with this discussion is that you’ve zeroed in on the significant findings and given very little consideration to your mostly null county-level associations. Usually, I think that’s fine, but in this case, I worry a reader may consider it fishing for a county-level association. Is it possible that the county-level variables are measured at too wide of a geographic level, e.g., would a zip-code level exposure be better?

We recognize the reviewer’s concern, and have updated the methods to make it clearer that the county-level measures chosen were purposefully selected to represent important social determinants of health. We also explain obstacles to conducting a ZIP-Code-level analysis (Reviewer 2 Comments 4 and 8). Both of these responses are repeated below for ease of reference.

County was chosen as the unit of geographic analysis for two reasons. First, some of the geographic measures were available at the county but not ZIP-Code-level. Second, due to data privacy policies, Tennessee Department of Health could not share patient-level data for regions with fewer than 5 HIV cases, hence using a smaller geographic area would have created more missing data in the analysis. We have updated the methods (Page 3, Lines 94-95), and described this as a limitation in the Discussion (Page 15, Lines 345-347).

Methods

Grounded in the social ecological model as a framework to consider barriers to linkage to HIV care, we assessed county-level community and structural factors representing healthcare access, socioeconomic status (SES) and disease burden.12,13 County was chosen as the unit of measurement because all of the variables of interest were measured or aggregated at this level. The measures were drawn from several sources including a CDC-developed Vulnerability Index (VI) which has helped to identify counties at high risk for incident HIV/HCV cases.14,15 The VI is comprised of measures such as percent of the population with a car, below the federal poverty level, who are White, have poor or fair health, are smokers, or have a disability. The VI also assesses per capita income, teen birth rate and HIV prevalence. These measures all represent social determinants that may pose barriers to linkage to care for HIV [Table 1].15 We retained all 15 variables from the CDC study and included 63 additional collected from the 2010 US Census, as well as TN state-specific indicators from the CDC and TDH surveillance data [Table 2]. Individuals were assigned exposure status to county-level factors based on county of residence. Additionally, counties with fewer than five PLWH were suppressed by the TDH according to data privacy regulations.

Discussion

Finally, more granular spatial analysis (ie zip code rather than county-level) was limited by both data suppression requirements for TDH, and lack of available public health data at the zip-code level.

County-level variables were purposefully selected to reflect healthcare access, community socioeconomic status, and community disease burden. We also included all 15 variables in the CDC-developed Vulnerability Index (VI) which has helped to identify counties at high risk for incident HIV/HCV cases. We hypothesized that these variables would be associated not only with new HIV infection, but also with linkage toHIV care. All variables included reflect important social determinants of health. We have edited the methods (Page 3, Lines 94-95) to clarify this.

Grounded in the social ecological model as a framework to consider barriers to linkage to HIV care, we assessed county-level community and structural factors representing important social determinants of health including healthcare access, socioeconomic status (SES) and disease burden.

10. The inpatient finding seems artifactual, given persons diagnosed while at an inpatient facility are admitted / on-site. I might re-frame this to look at outpatient as your reference. It seems more actionable to know if the health department or correctional facilities perform as well as outpatient facilities (table 2 suggests they perform worse).

We have updated the analysis to make the outpatient facility the reference group. The updated findings are presented in the Results section (Page 8, Lines 207-210), and Table 2.We agree that this finding could be artifactual depending on the reason for admission, and hospital practices. We have included this rationale in the discussion (Page 13, Lines 279-283).

Results

Compared to an outpatient facility or emergency room, patients diagnosed at inpatient facilities (aRR 1.18, 95%CI 1.08-1.29) were more likely and patients diagnosed at health departments or STD clinics (aRR 0.73, 95%CI 0.66-0.81) and correctional facilities (aRR 0.59, 95%CI 0.46-0.76) were less likely to establish HIV care [Table 2].

Discussion

Higher likelihood of linkage to HIV care from inpatient facilities may also reflect the fact that patients diagnosed in these settings are more ill, and thus will more readily establish care after hospitalization. Higher linkage to care from inpatient facilities could alternatively be artifactual – reflecting routine disease staging with CD4 count and viral load aseessment after diagnosis, and not in fact, linkage to care. Nonetheless, linkage to care from both inpatient facilities and outpatient facilities was lowest in Shelby County, the County seat of Memphis, and TN’s only priority county nationally targeted for EtE activities. Such findings represent an opportunity for improvement via optimization of linkage referrals at such sites, and implementation of models such as rapid treatment initiation to promote earlier linkage to care.27

11. Why declining linkage by 2016? What was happening or stopped happening – any TN DOH initiatives?

We have reviewed these data with our partners at TDH. The reasons for declining linkage in 2016 are likely multifactorial and could have included changes in lab reporting and surveillance practices especially with testing partners. As this was beyond the scope of our analysis, we have not opined on this specifically in the discussion.

12. Minor / editorial, but I would suggest using a word other than “incited” to highlight the increase in discussions about racial disparities (discussion, paragraph 2). Is it possible to cite HIV literature on structural determinants versus COVID?

We have changed this word as suggested in the Discussion (Page 15, Line 318).

These trends have furthered important discussions about systemic racial disparities in the US healthcare system, and may afford a critical opportunity to seriously consider how to address the structural factors driving such disparities in our healthcare system.

13. The structural racism angle/paragraph requires additional work. Your paper is about individual and community exposures. How do you tie them to structural racism (as a root cause)? And largely, your community level measures had a null association with linkage. Does this support your theory of structural racism as a root cause? Do you consider these county measures proxies for structural racism?

We have reframed this section and reorganized the discussion to better address these concerns. Our paper was about individual and community exposures, but the county-level exposures do represent important social determinants of health which are accepted drivers of structural racism. Our manuscript does add to necessary efforts to operationalize measures of structural racism – a complex challenge. The discussion was updated to reflex this (Page 13, Lines 266-268, and Page 15, Lines 313-334).

Our analysis of patterns and predictors of linkage to HIV care in TN between 2012 and 2016 highlights unsettling trends. First, despite concerted efforts from TDH, CDC and local partners, timely linkage to HIV care among newly diagnosed PLWH in TN has not only failed to improve over time, but TN now trails the nation in linking PLWH to care. Second, unacceptable racial disparities in linkage to care persist, as Blacks remain much less likely to link to care than Whites – even after accounting for a wide range of individual and structural factors that often are drivers of poor healthcare access and engagement. Our analysis has contributed to the growing call to operationalize measures of structural racism impacting health outcomes by identifying some potential systematic and programmatic opportunities that could be areas for intervention to begin to change this trend in TN.20 At the same time, our analysis highlights the difficulty of measuring this factor. Indeed, while we assessed 29 county-level variables that all represent social determinants of health, few were statistically significant predicotrs of individual linkage to care.

It is well-documented that HIV disproportionately affects the Black community in the US at large – a disparity that has persisted decades since the start of the HIV epidemic.6,8,9 Our study findings add to the literature highlighting a critical need to adopt comprehensive strategies to measure drivers of persistent and pervasive racial disparities in HIV outcomes to guide improvement. Today, the life-changing pandemic caused by the novel SARS-coronavirus-2 has targeted a floodlight on the power of structural racism to undermine public health as whole, and to precipitate disparities in COVID-19 infection, hospitalization, and death.35,36 These trends have furthered important discussions about systemic racial disparities in the US healthcare system, and may afford a critical opportunity to seriously consider how to address the structural factors driving such disparities in our healthcare system. In our analysis, the fact that racial disparities persistent despite accounting for both indidivual and county-level characteristics speaks to the insidious and complex nature of structural racism. Some, like former president of the American Public Health Association, Dr. Camara Jones, have called on us to recognize structural racism as “a system of structuring opportunity and assigning value based on the social interpretation of how one looks;” and the root cause of all differences in any health outcome associated with race.37 As such, racism is an important social determinant of health that necessitates a structural intervention.37 Acknowledging these complex dynamics, several American cities have declared racism as a public health crisis and committed to put racial equity at the core of all city procedures to advocate for policies that improve health in communities of color.38.39 Other cities and counties have made similar declarations, but it is clear that they must be accompanied by novel structural approaches to effectively reduce these disparities, and ultimately end the HIV epidemic.40

14. Limitations. I would add multiple testing. Are county level variables too broad of an exposure?

We have addressed the concern about multiple testing in response to Reviewer 2 Comment 2. We have included multiple comparisons as a limitation as suggested by the reviewer.

Submitted filename: Linkage to Care Reviewer Comments.doc

Click here for additional data file.

17 Jan 2022

26 Jan 2022

Reviewer #1: Minor clarifications/comments:

Comment 1. Thank you to the authors for this revised manuscript. I agree with the updated definition of linkage to care, though the text describing the linkage definition could use slight clarification. Can the authors please add more specificity to what is meant by “the first CD4 or HIV-1 RNA test result… after diagnosis”? (line 77-78). I presume you included any CD4/viral load that was collected at least 1 day after the diagnosis – if this is correct, please revise to state “after the date of diagnosis”.

We have updated this text as suggested which now reads as below:

The outcome of interest was linkage to HIV care, which was defined as receipt of the first CD4 or HIV-1 RNA test result captured via TN’s enhanced HIV/AIDS reporting system (eHARS) after the date of diagnosis, and was assessed at 30, 60, and 90 days.

Comment 2. I disagree with the statement in the discussion that “increases in teen birth rate were associated with a strong, but statistically significant increase in linkage to HIV care.” I suggest changing strong to “small”, given that the aRR is 1.02 for the association between this teen pregnancy and linkage to care.

We agree with this sentiment and have made the suggested change. The updated text reads as below:

Interestingly, increases in teen birth rate were associated with a small, statistically significant increase in linkage to HIV care.

Comment 3: Please correct the following omissions/typographical errors and re-run spell check throughout.

a. aRR for poor mental health days in the abstract is not updated with the new analysis – please check this

b. Line 212: I believe emergency room/department should be grouped with inpatient facility (as it is in Table 2). I believe “outpatient facility or emergency room” is an error here – please revise.

c. Line 253: remove inserted text “facilities ranged from 43% to”

d. Line 277: remove comma after “In addition”

e. Line 328: ‘geographic’ is misspelled

Each of these omissions/typographical errors was addressed. Additional line editing for spelling/grammar was also completed.

Reviewer #2: Revised TN Linkage Study

Comment 1. Minor. How do you interpret the teen birth rate finding? Is this a positive association: increasing teen birth rate is associated with increasing linkage to care? If so, do you think this is the result of confounding? I see this is clarified in the discussion narrative. But may want a sentence in results.

We appreciate this comment. We have not included any interpretation of the findings in the results section, but have included potential confounding to this discussion as an explanation for this finding. The updated text is below:

The reasons for this are not entirely clear, but could reflect intense wrap around services for pregnant and peripartum women to prevent mother-to-child HIV transmission or could reflect potential confounding.

Comment 2. Minor. The authors say, in the discussion and abstract, that racial disparities persisted even when adjusting for county-level social determinants of health. However, the modelling approach in the methods (and in Table 2) doesn’t appear to adjust for structural factors, just individual factors… “adjusting for a priori selected individual level covariates in multivariable analysis that were available in the surveillance data and known to be associated with the outcome of interest, including year of and age at diagnosis, sex, race/ethnicity and HIV transmission risk factor.” Does the individual model adjust for county-level social determinants? If so, please add.

The individual model adjusts for individual-level determinants, while the county-level model adjusts for both individual and county-level determinants. We have updated the text in the Results section to more clearly present data from the county-level model that supports our assertion about racial disparities.

Also, in this model which adjusted for both individual and county-level factors, White and Hispanic individuals had an increased risk of 30-day linkage to care compared to Black individuals (aRR 1.33, 95%CI 1.30-1.37, aRR 1.44, 95% CI 1.41-1.47 respectively) [data not shown].

We have also updated Figure 2 (below) to reflect data from the county-level model (which adjusted for individual and county-level variables and included a county-by-race/ethnicity interaction term).

We analyzed the marginal probabilities of linkage at 30-days in the four highest-burden metropolitan counties by race/ethnicity and found that Black patients persistently had the lowest probability of 30-day linkage to care as compared to both White and Hispanic individuals when adjusting for individual level factors, and when adjusting for both individual and county-level factors when interacting individual-level race/ethnicity with county of residence [Figure 2]. Racial disparities were least prominent in Davidson County (the county seat of Nashville), which also had the highest marginal probability of linkage to care of the four highest-HIV-burdened counties.

Comment 3. Minor – line 328, don’t follow the euphemism ‘paints a picture’. I think you’re trying to say that segregation is correlated with geographic factors? Was it specifically correlated with poor mental health and therefore correlated with poor linkage? The point you’re trying to make about segregation is poorly constructed. Suggest revising.

We have updated the language to more clearly explain the likely relationship between racial segregation and county-level geographic factors. The updated text is in the Discussion and reads as below:

Additionally, the high correlation of residential racial segregation (White, non-White) with many county-level factors, while not surprising, further underscores the relationship between race and a range of geographic factors that can impact health.

Comment 4. Minor. Multiple testing is not specified in the limitations, although the authors suggest it was included. Imagine this was an oversight.

We have updated the discussion to include the limitations posed by multiple testing.

We acknowledge that we have included many covariates in our analyses, and could be subject to limitations from multiple testing.

Submitted filename: Response to Reviewers 1.26.22.doc

Click here for additional data file.

14 Feb 2022

Individual, Community, and Structural Factors Associated with Linkage to HIV Care Among People Diagnosed with HIV in Tennessee

PONE-D-21-13112R2

Dear Dr. Ahonkhai,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Natalie J. Shook

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

18 Feb 2022

PONE-D-21-13112R2

Individual, Community, and Structural Factors Associated with Linkage to HIV Care Among People Diagnosed with HIV in Tennessee

Dear Dr. Ahonkhai:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Natalie J. Shook

Academic Editor

PLOS ONE