Randomized Comparison of Universal and Targeted HIV Screening in the Emergency Department

Abstract

Objective

Universal HIV screening is recommended but challenging to implement. Selectively targeting those at risk is thought to miss cases, but prior studies are limited by narrow risk criteria, incomplete implementation, and absence of direct comparisons. We hypothesized that targeted HIV screening, when fully implemented and using maximally broad risk criteria, could detect nearly as many cases as universal screening with many fewer tests.

Methods

This single-center, cluster-randomized trial compared universal and targeted patient selection for HIV screening in a lower prevalence urban emergency department. Patients were excluded for age (<18, >64), known HIV infection, or prior approach for HIV testing that day. Targeted screening was offered for any risk indicator identified from charts, staff referral, or self-disclosure. Universal screening was offered regardless of risk. Baseline seroprevalence was estimated from consecutive de-identified blood samples.

Results

There were 9,572 eligible visits during which the patient was approached. For universal screening, 40.8% (1,915/4,692) consented with six newly diagnosed (0.31%, CI₉₅ 0.13%–0.65%). For targeted screening, 37% (1,813/4,880) had no testing indication. Of the 3,067 remaining, 1,454 (47.4%) consented with 3 newly diagnosed (0.22%, CI₉₅ 0.06%–0.55%). Estimated seroprevalence was 0.36% (CI₉₅ 0.16%–0.70%). Targeted screening had a higher proportion consenting (47.4% v. 40.8%, p<0.002), but a lower proportion of ED encounters with testing (29.7% v. 40.7%, p<0.002).

Conclusions

Targeted screening, even when fully implemented with maximally permissive selection, offered no important increase in positivity rate or decrease in tests performed. Universal screening diagnosed more cases, because more were tested, despite a modestly lower consent rate.

INTRODUCTION

Early diagnosis of HIV is critical.¹ Once diagnosed, infected individuals can engage in behavior change^2,3 and treatment^4–7 which combine to decrease infectivity and reduce cost. Emergency departments (EDs) have been emphasized as a partner in HIV screening because of high patient volume, access to vulnerable populations, and utilization by a broad spectrum of society.^1,8–20 Unfortunately, EDs are already overburdened,^21–23 and separate, dedicated funds to scale-up testing are not widely available. Identifying the most efficient approaches to screening will facilitate implementation.

Deciding which patients should be tested is a major challenge. Public health advocates have called for universal or non-targeted screening.^1,18,24 This requires many tests, and concerns persist about feasibility and effectiveness,^15,25–42 particularly when prevalence is low or presumed low. Selectively targeting increased likelihood of undiagnosed infection requires fewer tests, but failure to identify all cases is frequently cited.^1,18,43–48 However, targeted screening may have failed because of incomplete implementation rather than inadequately sensitive criteria.³⁵ If so, expanding targeting criteria beyond conventional risk criteria should increase sensitivity. There has been little comparative study about the trade-offs of such an approach or which precise targeting criteria would be most advantageous.

A growing number of studies have reported yield from non-targeted screening.^{17,20,36,37,39,49–59} Direct comparisons between non-targeted and targeted approaches are scarce ^17,20,54 and confounded by many other factors such as consent method, assay type, targeting criteria and assessment method, who conducts screening, and incomplete implementation. We compared the efficacy of universal and targeted screening in the context of an opt-in ED HIV screening program, hypothesizing that given full implementation and maximally permissive selection criteria, targeted screening would detect nearly as many cases while requiring many fewer tests.

METHODS

Design Overview

This single-center, cluster-randomized trial compared the yield of universal and targeted screening for HIV in the ED, in the context of two concurrent seroprevalence studies. Seroprevalence estimates were obtained either by approaching patients consecutively (patient-based) or from de-identified, discarded blood samples, originally obtained for clinical purposes (remnant-based).

The study enrolled patients from January 2008 through December 2010. Blocks of patients were randomized to be approached for HIV screening on a universal or targeted basis or, for the first two years of enrollment, in the patient-based seroprevalence study. The remnant-based seroprevalence study was conducted over the third year of the screening study. The study was approved by the Institutional Review Board and overseen by a Data and Safety Monitoring Board (DSMB). No harms or unintended study effects were recorded.

Setting and Participants

This study was conducted in the ED of a Midwestern, urban, 450-bed, teaching hospital with 90,000 annual ED patient encounters. Of ED patients, about 50% were black, 0.5% were Hispanic, and 40% were uninsured. Pediatric patients were rarely seen, since a large pediatric ED is located nearby. A publicly funded HIV counseling and testing program has operated in this ED since 1998.^60–62 The cumulative diagnosis rate of HIV/AIDS in the surrounding county is 229 per 100,000 persons.⁶³

For the screening arms, patients were ineligible for age <18 or >64, known HIV infection, or previous approach for ED HIV testing that day. Post-exposure testing (e.g. occupational, assault,) was not provided. For the seroprevalence arms, patients were ineligible for age <18 or >64.

Randomization and Interventions

Randomization

Patients were approached for universal screening, targeted screening, or the patient-based seroprevalence estimate according to cluster randomization. Clusters were defined by combinations of four 6 hour periods spanning the full 24 hours and four separated ED patient care areas. The probability of each combination being a study period was set to achieve about 4 study periods per week. Study periods were then randomly assigned to being one of the three study arms.

Study periods were cancelled without replacement if: 1) the assigned ED area was closed during that time, 2) two study periods were adjacent in time and space, 3) staffing was insufficient. Patients were allocated to periods based on their physical presence and not their time of arrival to the ED.

ED HIV Counseling and Testing Program

The screening study used the same procedures as the existing clinical screening program,^60–62 but staffing was augmented to include every eligible patient. Additional study intervention was limited to randomizing patients between different selection criteria, both of which were used regularly by the program prior to the study.

HIV testing was offered at no cost with signed, opt-in consent, detailed risk-assessment, and formal prevention counseling. Ohio statutes and hospital policy mandated opt-in consent for HIV testing until 2011. There was no separate research consent; the clinical consent for HIV testing included IRB approved language advising that information could be “analyzed to monitor the quality of the testing program and combined and reported to improve our understanding of issues related to the spread of HIV”.

Services were provided by cadre of 10 to 12 counselors dedicated counselors operating in parallel with usual ED care. Counselors worked between 16 and 40 hours per week for about a year or more. They were trained in basic information regarding HIV illness, testing, and transmission, formal prevention counseling, and ED and study operations. Training included 12 hours of didactic instruction, followed by 4 hours of role-play and observation or initial patient encounters. This was followed with periodic observation of counselor performance, feedback based on quality assurance review of data forms and enrollment logs, monthly meetings, and remediation if needed. Counselor work schedules were created entirely separately from activity randomization, so there would be no systematic difference in the way individual counselors were allocated between study arms. Risk assessment and counseling used a questionnaire driven interview to promote an individualized plan for risk reduction. The questionnaire, recommendations, and HIV testing information were entered into the program’s electronic record system.

Before October 2008, testing used conventional HIV ELISA, with confirmatory Western blot. In 2008, testing switched to rapid assay (OraQuick ADVANCE^® Rapid HIV-1/2 Antibody Test using oral fluid). If reactive, the patient underwent repeat rapid assay but with a whole blood sample and also had blood drawn for confirmatory testing by Western blot. Negative results were reported to patients by telephone (conventional assay) or in person (rapid assay); positive results were provided in person.

Selection and Recruitment for HIV Screening

Universal Arm

Counselors approached every patient not known to meet exclusion criteria. Counselors could encourage participation by discussing the importance of testing generally, but did not use individualized risk information to motivate testing.

Targeted Arm

Counselors reviewed triage notations and electronic medical records to target patients, or acted on staff referral. Indications for targeting contained over fifty items (supplemental electronic appendix), including: 1) clinician identified signs and symptoms of HIV, and 2) clinician or counselor identified risk behaviors, homelessness, mental illness, STD exposure or infection, violence, substance use, pregnancy, and incarceration. Counselors did not systematically assess patients for every possible risk, but were well familiar with criteria and approached patients whenever at least one risk was apparent. Patients for whom no risk was readily apparent were asked directly if they had 1) ever injected drugs, exchanged sex for drugs or money, had sex with a man (if male), or had sex with a partner with or at-risk for HIV, or 2) in the past two years used cocaine or methamphetamine, had sex while using drugs or alcohol, been diagnosed with a STD, or had more than one sex partner. Counselors could use risk information to encourage testing.

In all cases, counselors recorded the reasons prompting the test offer. Inability to complete the testing offer was counted as a failed approach, separate from declined offers. Counselors were necessarily unblinded, though separation of study arms was enforced through training, oversight, and color-coding of study forms. Patients may have been aware of indications for testing but not that these varied systematically.

Recruitment for Seroprevalence Estimation

Patient-Based Seroprevalence

Study personnel consecutively approached every eligible patient, to invite participation in a “study of diseases of public health importance”. Patients received $10 for a blood sample and $5 a health history. The consent process emphasized that data would be stripped of all identifiers before any analysis, and disclosed HIV as one disease among others for which samples might be tested.

Remnant-Based Seroprevalence

Discarded blood samples were obtained from the hospital laboratory for ED patients one week after receiving care during one of seventeen 24-hour periods. Periods were purposively selected to provide data for one or two days each month and all days of the week. Research consent requirements were waived by the IRB.

Data collection

HIV counseling and testing data were extracted from screening program records. For the patient-based seroprevalence study, health questionnaires included information about HIV risk integrated within a broad health history. For the remnant-based seroprevalence estimate, trained abstractors conducted a structured chart review to collect analogous information. HIV status for both seroprevalence components was determined by a sequential method. If a clinical HIV test was performed and negative on or after the date of enrollment, the seroprevalence sample was presumed antibody negative without assay. Remaining samples were assayed in pools of 100 µL serum samples from ten subjects, created from constituent pools of five.^64–66 Pools were tested for HIV antibodies by standard immunoassay (Abbott HIVAB™ HIV-1/HIV-2 (rDNA) EIA) and OraQuick ADVANCE^® Rapid HIV-1/2 Antibody Test. Individual positive samples were tested a second time and then confirmed by standard Western blot.

Outcomes and Follow-up

The primary outcome was the proportion of new HIV diagnoses in the targeted and universal study arm. A positive HIV test was assumed to be a new diagnosis when there was no indication of prior HIV diagnosis from other sources such as the patient, medical record, other treatment providers, or health department.

Secondary outcomes included: proportion of eligible and approachable ED patients who were tested and who were newly diagnosed; proportion offered testing who consented; risk-profile of tested patients; proportion tested who were notified; the number of positive patients linked to care; reasons for declining testing; and initial CD4 count of newly diagnosed patients. Seroprevalence estimates are provided to contextualize the screening comparison.

Statistical Analysis

Comparison between Screening Arms

The proportion positive, or testing yield, was compared by estimating the difference between proportions and the 95% CI of the differences. Secondary comparisons between groups used the Chi-square test, Fisher’s Exact test, or the Mann Whitney U test. All statistical analyses were conducted using SPSS 20.0 (IBM Corporation, Armonk, NY). We did not adjust for clustering in our analysis; there was no reason to expect that patients within a cluster were more similar than patients in different clusters.

Individual patients often visit an ED more than once.⁶⁷ We used encounters rather than patients as our primary unit of analysis.⁶⁸ Because we have previously shown some differences in HIV screening program outcomes depending on unit of analysis,⁶⁹ we conducted a secondary analysis to determine whether using only the first or the last test for patients with multiple encounters would influence the results.

Seroprevalence Estimation

The patient-based and remnant-based seroprevalence studies were combined for analysis, although proportion positive was also calculated for each separately. Only the final enrollment was included in cases of duplicate enrollment. Patients were excluded from analysis if already diagnosed with HIV, either by self-report or medical record, or if HIV status could not be determined (e.g. sample insufficient).

Sample size Considerations

The positivity rate using targeted screening was expected to be approximately 1%. We selected 0.5% as the clinically important absolute difference in positivity that would be sufficient to influence program planning and design. A sample size of 2000 completed tests in each group was required for 80% power to detect a two-sided 95% CI for the difference in proportion positive to be ±0.5%. The study was stopped before reaching the enrollment goal because of exhausted resources.

RESULTS

Study Sample

Comparison of Patient Selection Methods

Of the 666 study periods specified by randomization, 600 were completed. In 58 cases the ED care area was closed or study periods were adjacent; staffing was insufficient in 8 cases. Figure 1 illustrates study flow. Of the 13,186 patient encounters, 681 (5%) were missed by counselors, 1,777 were ineligible, and 1,156 could not be approached due to patient or clinical care reasons, such as impaired cognition or critical illness. Missed patients were evenly distributed between study arms, and 63% were missed because the patient left the ED before counselors had a chance to interact with them. In the universal screening arm, 1,915/ 4,692 (40.8%, CI₉₅ 39.4%–42.2%) consented. When targeting, 1,813/4,880 (37.2%, CI₉₅ 35.8%–38.5%) had no apparent testing indication. Table 1 lists reasons for targeting the remaining 3,067 who were offered testing. There were 1,454/3,067 targeted patients who consented to testing (47.4%, CI₉₅ 45.6%–49.2%).

Figure 1.

Flow of Patients Through the Study

Table 1.

Reason for Offer of Targeted Testing^*

	Total		Refused		Consented
	N	(%)	N	(%)	N	(%)
Counselor identified risks	2,091	(68.2)	1,149	(71.2)	942	(64.8)
Systematic high-risk screen	818	(26.7)	415	(25.7)	403	(27.7)
Not documented	61	(2.0)	37	(2.3)	24	(1.7)
Self - request	60	(2.0)	0	(0.0)	60	(4.1)
Clinical staff identified risks	23	(0.7)	9	(0.6)	14	(1.0)
Clinical staff identified symptoms	14	(0.5)	3	(0.2)	11	(0.8)

^*The reason for offering testing in the universal arm was, by definition, patient presence in the ED.

Seroprevalence Estimate

For the patient-based seroprevalence estimate, 1,934 were eligible for enrollment. The initial approach could not be completed for 334 (17.3%), 112 (5.8%) were missed, and 454 (23.5%) declined. Of the 1,034 who consented, 24 (2.3%) were already diagnosed with HIV and 37 (3.3%) would have been duplicate enrollments. There was insufficient sample to determine serostatus in 45 (4.6%) who did not have a subsequent negative test documented in the medical record. Two were inadvertently assigned the same sample identification number and were excluded. For the sample-based seroprevalence estimate, there were 1,083 samples collected for patients aged 18 to 64 years. Of these, 22 (2.0%) were previously diagnosed with HIV, 29 (2.7%) were duplicate enrollments, and 10 (0.9%) had insufficient sample and no subsequent record of negative HIV test result in the clinical record.

Primary Outcome

Among the 1,911 patients tested on a universal basis, 6 were newly diagnosed (0.31%, CI₉₅ 0.13%–0.65%). For the 1,451 tested on a targeted basis, 3 were newly diagnosed (0.22%, CI₉₅ 0.06%–0.55%). In the corresponding combined seroprevalence study, 7/1,948 (0.36%, CI₉₅ 0.16%–0.70%) were found to be HIV antibody positive and not previously known to be diagnosed with HIV (4/926 in the patient-based component and 3/1,022 in the remnant-based component).

Secondary Outcomes

Characteristics of Tested Patients (Table 2)

Table 2.

Patient Characteristics Stratified by Study Arm

		Targeted N=1,451		Universal N=1,911		Seroprevalence* N=1,948
		N	(%)	N	(%)	N	(%)
Age	18–29	559	(38.5)	685	(35.8)	497	(25.5)
	30–39	300	(20.7)	384	(20.1)	360	(18.5)
	40–49	339	(23.4)	444	(23.2)	447	(22.9)
	50–64	256	(17.6)	402	(21.0)	608	(31.2)
	Not documented	--	--	--	--	36	(1.8)
Sex^	Male	736	(50.7)	951	(49.8)	955	(49.9)
	Female	713	(49.1)	955	(50.0)	963	(49.4)
	Not documented	4	(0.3)	9	(0.5)	30	(1.5)
	Transgendered	1	(0.1)	0	(0.0)	0	(0.0)
Race	White	530	(36.5)	716	(37.5)	820	(42.1)
	Black	886	(61.1)	1131	(59.2)	990	(50.8)
	Hispanic	23	(1.6)	46	(2.4)	0	(0.0)
	Other/Not documented	15	(1.0)	22	(1.2)	138	(7.1)
Last HIV test	Never	362	(24.9)	548	(28.7)	199	(10.2)
	Last year	364	(25.1)	446	(23.3)	323	(16.6)
	Over one year ago	676	(46.6)	868	(45.4)	397	(20.4)
	Not documented	52	(3.6)	53	(2.8)	1029	(52.8)
Multiple partners, not always using condoms#		618	(42.6)	647	(33.9)	243	(12.5)
Person ever
having vaginal sex		1,253	(86.4)	1,622	(84.9)	899	(46.1)
having any anal sex		323	(22.3)	351	(18.4)	175	(9.0)
receiving anal sex		167	(11.5)	193	(10.1)	108	(5.5)
giving anal sex		188	(13.0)	187	(9.8)	83	(4.3)
having sex with an at-risk partner		509	(35.1)	544	(28.5)	291	(14.9)
man having sex with men		49	(6.7)	50	(2.6)	20	(2.1)
having sex with a prisoner		377	(26.0)	397	(20.8)	201	(10.3)
having sex with an iv drug user		137	(9.4)	146	(7.6)	72	(3.7)
having sex with a std positive partner		89	(6.1)	90	(4.7)	70	(3.6)
having sex with an HIV positive partner		20	(1.4)	16	(0.8)	2	(0.1)
having sex under the influence		841	(57.8)	953	(49.9)	558	(28.6)
trading drugs or money for sex		146	(10.1)	147	(7.7)	77	(4.0)
had a prior std		741	(51.1)	852	(44.6)	446	(22.9)
used injection drugs		131	(9.0)	134	(7.0)	121	(6.2)
ever used crack&		311	(21.4)	287	(15.0)	329	(16.9)

Missing data was coded as “none reported”. Approximately 8% of risk data (sexual and other risk) for the targeted and universal arms was missing and in the seroprevalence arm approximately 1% of the risk data was missing.

^*Combined seroprevalence estimate (prospective consecutive enrollment + remnant sample collection)

^{^}Sex as perceived by patient

^#Within the past year

^&defined as non-injected cocaine use in seroprevolence study

Screening groups were of similar demographics and self-reported prior testing history. Patients in the targeted arm self-reported risk behavior with greater frequency. CD4 was available for 5 patients of the 9 new diagnoses identified by screening (154, 256, 493, 512, and 882 cells/mm³); four of these were obtained within six months of diagnosis.

Consent and Reasons for Declining Testing (Table 3)

Table 3.

Reasons for Declining Offered Testing by Study Arm

		Targeted		Universal
		N	(%)	N	(%)
Confidentiality
Prefers anonymous testing		8	0.5	21	(0.8)
Setting
Doesn't want to interview/counseling		414	(25.7)	697	(25.1)
Doesn't want to wait for ED testing		67	(4.2)	115	(4.1)
Problems with followup notification*		0	(0.0)	1	(0.0)
Doesn't want venipuncture*		15	(0.9)	20	(0.7)
Wants rapid testing*		11	(0.7)	8	(0.3)
Privacy of environment insufficient		3	(0.2)	5	(0.2)
Prior testing
Negative test < 3 months prior		528	(32.7)	758	(27.3)
Negative test > 3 months prior		382	(23.7)	536	(19.3)
Test in last month & results pending		17	(1.1)	28	(1.0)
Blood donor < 3 months prior		10	(0.6)	16	(0.6)
Blood donor > 3 months prior		2	(0.1)	6	(0.2)
Deny risk&
abstains from sex		120	(7.4)	278	(10.0)
abstains from IDU		60	(3.7)	115	(4.1)
Monogamous		235	(14.6)	491	(17.7)
uses barrier methods		17	(1.1)	28	(1.0)
uses clean needles		2	(0.1)	2	(0.1)
Other reasons
Too much pain/nausea/illness		255	(15.8)	432	(15.6)
Alternative test planned		44	(2.7)	58	(2.1)
Afraid/doesn’t want to know results		28	(1.7)	37	(1.3)
HIV positive
Already diagnosed HIV positive		7	(0.4)	16	0.6)

Patients could indicate more than one reason for declining testing. Reasons were coded prospectively according to these a priori categories..

^*Relevant for early portion of testing program operating with conventional, non-rapid, testing

^&Reasons why the patient thought themselves not to be at risk

More patients consented when approached on a targeted basis than a universal basis (1,454/3,067 (47.4%) v. 1,915/4,692 (40.8%); p<0.002). However, the proportion of all ED patients who were approached and tested was greater for the universal than the targeted arm (1,911/4,692 (40.7%) v. 1,454/3,067 (29.7%); p<0.002). When compared with patients who declined testing in the universal arm, patients who declined testing in the targeted arm more often reported their reason as prior negative testing (59.8% v. 48.9%; p<0.001) and less often that they were not at risk (23.1% v. 28.8%; p<0.001).

Repeat Testing

Of the 3,369 HIV tests conducted, 142 (4%) were for patients that had been previously tested in the study (83 universal; 59 targeted). The median duration of time between repeat tests was 267 days (range 7–1,024 days). Overall, 3,107 (96%) were tested once, 103 (3%) were tested twice, and 17 (0.5%) were tested 3 or more times. Sensitivity analysis including only the first or the last encounter had no effect on results.

DISCUSSION

The CDC’s rationale for recommending universal screening is strong,¹ but implementation remains a challenge. We tested the hypothesis that a broadly permissive and fully implemented targeted screening strategy would require considerably fewer tests than a universal approach yet achieve a similar number of new diagnoses. We found that this targeted screening strategy was unlikely to be different from universal screening in any clinically significant way. Reduction in testing volume when targeting was matched proportionally by a reduction in the detection of new cases.

Our study offers two primary advances. First, we applied sufficient resources to achieve comprehensive implementation. Prior studies have not delineated whether targeting was inadequately sensitive or inadequately applied. Second, we used a broad set of targeting criteria, whereby any indication more specific than simple presence in the ED led to an offer. Combined, these factors 1) gave targeting the best chance of detecting as many cases as universal screening, while still avoiding testing where no indication of any risk was apparent, and 2) mimicked our existing screening program, which tries to balance the need for manageable testing volumes with the need to detect all cases.

We propose several explanations for the unexpected similarity between targeted and universal screening in our study. Although our environment has low HIV prevalence, risk for HIV was endemic throughout both study arms. We also posit that the lower risk population expected in the universal arm may have declined testing. In addition, targeting criteria were so broad that when fully implemented, the targeted group was not very different from the universal group. Of note, partial implementation of the same targeting criteria by our clinical screening program appears much more selective. The positivity rate of the clinical program was close to 1% both immediately before and immediately after this study (data not shown). This suggests counselors naturally select higher risk patients when resources are insufficient to test all those with some identified risk, even when eligibility criteria are the same. If true, this finding highlights the potential for differences between efficacy and effectiveness studies.

Consent rate was higher in the targeted arm than the universal arm, despite the use of signed, opt-in consent in both arms. This is most likely because targeted patients had higher behavioral risk,^70,71 but it is possible that a lesser emphasis on the need for testing in the universal arm contributed to the lower consent rate. We think this is unlikely, counselors were encouraged to emphasize the need for testing in both arms, and using risk information to motivate testing has not been highly efficacious.^72,73

The finding that consent varies by patient selection strategy demonstrates the importance of controlled comparisons. Outcomes from studies that alter or fail to control for such operational features will be difficult to interpret. Not only do multiple variables influence program outcomes, but these variables interact with one another in unknown ways. Interactions between different aspects of a screening likely extend beyond consent and patient selection to include myriad factors such as assay method, which staff offer testing, and when in the course of care testing is offered.

Recommendations to expand HIV testing have encouraged changes in methods for both patient selection and consent. Our study provides some insight into the relative impact and interaction of those factors. The proportion who declined universal screening (59%) was greater than the proportion that were not targeted (37%). Still, non-targeted approaches may result in greater increases in testing than changes in consent strategy. The number without perceived risk would be much greater than in our study if using only conventional behavioral risk factors, as is common. Also, improvements in consent are likely to be suboptimal in most settings. Nonetheless, research to improve consent rates should remain a high priority, particularly given the widely ranging rates observed with “opt-out” strategies (e.g. 24%–99.7%).^17,59 Our data also indicate that strategies to increase consent are particularly important for non-targeted selection, since consent was less frequent in the universal arm.

Consent may be even more important when going beyond the number tested to consider the number of new infections detected. Extrapolation from our seroprevalence estimate of undiagnosed HIV (0.4%) would suggest 17 total cases among those eligible and approachable in the universal arm. Only six were identified by screening. Moreover, this extrapolation assumes an identical proportion with undiagnosed HIV among those who were and were not tested, while others have suggested positivity rates to be greater among those who decline testing than among those who consent.^74,75

Our findings inform but do not resolve the tension between universal and targeted patient selection strategies. Methods to optimize the trade-off between missed cases and fewer tests when resources are insufficient remain largely unexplored. Recent reports have suggested the Denver HIV Risk Score to be a promising instrument to enhance targeting.^76,77 This and other targeting approaches, including our baseline clinical program, should be compared individually with universal screening.

Our study should be considered in light of several limitations. Our results may not be generalizable to centers with different epidemiology; EDs differ in terms of disease prevalence even in the same region,^78,79 and EDs are certain to differ from other healthcare venues.^10,19 Our study was strengthened by controlling overall screening methodology to isolate differences resulting from patient selection strategy. Nonetheless, overall results were necessarily influenced by our screening model, which is only one of many.^18,80 A tiny proportion of those tested in the targeted screening arm were actually referred for diagnostic testing¹⁸ (i.e. testing for signs and symptoms of HIV). The number of screened patients who may also have met criteria for diagnostic testing is unknown. Our targeting criteria were not used as an instrument to systematically assess risk. Our analysis does not consider the relative costs of the two patient selection strategies. Prospective seroprevalence sampling was biased by the consent requirement, but included patients who would not have had blood remnants available and used survey for health history. The remnant study was not biased by consent, but only included individuals with adequate remnant samples, and health history was obtained by chart review. The study was stopped before planned sample size was reached, although it is inconceivable that further testing could have led to a difference in proportion positive greater than ±0.5%.

When using opt-in consent in low prevalence areas, universal HIV screening results in more testing with a similar proportion positive compared to a fully implemented targeting strategy using broad eligibility. More specific criteria would be needed whenever a higher positivity rate or reduced testing volumes are desired. Increasing consent rate may be as important as patient selection strategy to increase testing and is particularly necessary for non-targeted approaches.