Abstract
Objective
We used data from a routine HIV testing program to develop risk scores to identify patients with undiagnosed HIV infection while reducing the number of total tests performed.
Design
Multivariate logistic regression
Methods
We included demographic factors from HIV testing data collected in 134 Botswana Ministry of Health & Wellness facilities during 2 periods (10/1/2018–8/19/2019 & 12/1/2019–3/30/2020). In period 2, the program collected additional demographic and risk factors. We randomly split each period into prediction/validation datasets and used multivariate logistic regression to identify factors associated with positivity; factors with adjusted odds ratios ≥1.5 were included in the risk score with weights equal to their coefficient. We applied a range of risk score cutoffs to validation datasets to determine tests averted, test positivity, positives missed, and costs averted.
Results
In period 1, 3 factors were significantly associated with HIV positivity (coefficients range 0.44–0.87). In period 2, 12 such factors were identified (coefficients range 0.44–1.37). In Period 1, application of risk score cutoff ≥1.0 would result in 50% fewer tests performed and capture 61% of positives. In Period 2, a cutoff ≥1.0 would result in 13% fewer tests and capture 96% of positives; a cutoff ≥2.0 would result in 40% fewer tests and capture 83% of positives. Costs averted ranged from 12.1–52.3%.
Conclusion
Botswana’s testing program could decrease testing volume but may delay diagnosis of some positive patients. Whether this trade-off is worthwhile depends on operational considerations, impact of testing volume on program costs, and implications of delayed diagnoses.
Keywords: HIV infections, HIV testing, HIV seropositivity, Botswana, Risk Factors
Introduction
Botswana has the world’s third highest adult prevalence of HIV infection. Of the approximately 2.3 million adults, over 20% population are estimated to be living with HIV and an additional 0.5% (11,500) are newly diagnosed annually1. Botswana was one of seven sub-Saharan African countries to meet the UNAIDS fast-track 90-90-90 by 2020 goals1 and has made substantial progress towards the 2030 UNAIDS 95-95-95 goals. As of 2019, an estimated 92% of the 340,000–410,000 persons living with HIV knew their status, while 89% of persons who knew their status are on treatment, and 96% of persons who knew their HIV status and are on treatment have achieved viral suppression1; 79% of all persons living with HIV, diagnosed and undiagnosed, are thought to be virally suppressed1. Botswana has made remarkable progress in the fight against HIV, but these data demonstrate that a sizable population of persons living with HIV still remain undiagnosed. Diagnosing this population of persons living with HIV and linking those persons to effective treatment is paramount to reducing further transmission.
Starting in 2017, the US President’s Emergency Plan for AIDS Relief (PEPFAR) began an effort to improve the efficiency of HIV testing by decreasing testing volume and focusing testing on populations with high HIV positivity2. Ideally, this effort would sustain HIV case-finding while decreasing program costs. Several previous analyses from sub-Saharan African settings have examined the use of demographic and clinical information for targeting resource-intensive tests for identification of acute HIV infection3,4. However, few data exist on how programs can optimally focus testing for increasing positivity to include identification of chronic, previously undiagnosed infections. We sought to use data collected as part of Botswana’s Ministry of Health and Wellness (MOHW) routine HIV testing program to develop risk scores that would identify the majority of patients with undiagnosed HIV infection while allowing healthcare facilities to reduce the number of tests performed.
Methods
The study used retrospective data collected as part of HIV testing procedures during two distinct periods between October 1, 2018-March 30, 2020 in the 134 MOHW facilities supported by the University of Washington International Training and Education Center for Health (I-TECH). Period one comprised data collected from October 2018-August 2019 and included the following variables characterizing persons being tested: sex at birth, age, citizenship, the testing point within the healthcare facility at which testing occurred (e.g. outpatient clinic, antenatal care) and the testing strategy under which the test was undertaken (antenatal care/Prevention of mother-to-child transmission (PMTCT) program, emergency, partner services, inpatient, pediatric malnutrition, presumptive/confirmed Tuberculosis (TB), sexually transmitted infection (STI), voluntary counseling and testing (VCT), and other provider initiated testing and counseling). Testing “mobilisers” working in participating healthcare facilities identified clients screened and identified patients eligible for HIV testing. Eligible patients were ushered by the mobilisers to site-based Health Care Auxiliaries, who were trained and certified by the MOHW to provide HIV counseling and testing.
In an effort to better identify patient characteristics associated with HIV positivity, in December 2019 the program began collecting the following additional demographic and risk factor information from all persons tested for HIV: marital status, educational attainment, employment status, TB symptoms (cough, night sweats, or fever for a duration of ≥2 weeks, recent unexplained weight loss), HIV symptoms (oral thrush, herpes zoster, or swollen lymph nodes), symptoms of sexually transmitted infection (STI) (dysuria, urethral or vaginal discharge), history of STI diagnosis or treatment in the prior year, time since last HIV test (never, 3–6 months ago, 6–12 months ago, or >1 year ago; those testing <3 months prior were not eligible for testing), and behavioral risks (condomless sex, condomless sex with a partner known to be living with HIV, sex under the influence of alcohol) since the last HIV test. Data were regularly transferred to I-TECH headquarters in Gaborone, where a dedicated team performed data cleaning. In period 2, data were deduplicated by matching identified positive cases with the HIV treatment initiation registers at the facility and validation against the national HIV data warehouse; any persons previously diagnosed were removed from the dataset. In period 2, a total of 26 duplicate positive cases (~1.2% of total positive cases in period 2) were identified and removed from the dataset. Records with missing variable values were retained in the dataset.
We randomly split each period’s data into prediction and validation datasets of equal size. We then used multivariate logistic regression to identify factors (e.g., demographic characteristics, testing strategies) associated with HIV positivity within each period. Records from August 20, 2019 - November 30, 2019 were excluded as sites were piloting the new screening tool with expanded risk factor collection during this period. Factors with significant adjusted odds ratios (aOR) ≥1.5 at α=0.95 were included as items in the risk score for each period with a weight equal to their regression coefficient5. Testing strategies and sectors where I-TECH staff believed that cessation of testing would be unacceptable a priori (patients with presumptive/confirmed tuberculosis or STIs, those tested as part of voluntary counseling and testing or partner services, and patients in the antenatal, labor & delivery, pediatrics and gynecology departments) were excluded from model development and included in both sets of testing criteria regardless of risk score. We applied the new testing criteria to each period’s validation dataset to compare the number of tests performed, test positivity, proportion of positives missed, positivity among those that would not be tested, and costs that would have been averted if the criteria had been used to determine whether HIV testing was indicated. In period 1, we used a score cutoff of ≥1.0; in period 2, we examined performance of score cutoffs ≥1.0, ≥1.5, and ≥2.0. We also applied the risk score components and cutoff developed using period 1 data to period 2’s validation dataset to compare its performance against scores incorporating the enhanced demographic and risk factor information.
Costs were derived based on total budgets allocated to I-TECH to support HIV testing and estimated costs of test procurement. HIV test kits are procured by the MOHW. The estimated cost per negative test was US$2 (Determine™ HIV-1/2). The estimated cost per positive test was US$8.50, which includes costs associated with the Determine™ HIV test in addition to a confirmatory test (Trinity Biotech Uni-Gold™ Recombignen™ HIV-1/2).
In addition to the HIV test kits and supplies, HIV testing service program implementation costs per test were estimated to be US$16.23/test, which includes staff salaries (HCAs, mobilizers, program officers, M&E staff, support staff), technical assistance and support site-visit expenses, including travel and per diem, performance-based incentives rewards, administrative expenses such as rent, utilities, printing and stationery, demand creation activities, equipment and supplies such as computers, laptops, tablets for data collection, internet data bundles, printers, and cartridges. This cost was derived by dividing the total testing program expenditures ($US4,934,687) by the number of HIV tests (304,070) performed during the study periods.
Because administrative costs would likely not decrease proportionally with a reduction in tests performed, we estimated the reduction in costs with an assumed 25%, 50%, and 75% reduction in this administrative cost per test.
Results
Study Population
Staff performed a total of 262,230 and 41,840 tests during periods 1 and 2 respectively (Table 1). Compared to period 1, a higher proportion of tests in period 2 were administered to citizens of countries other than Botswana and a lower proportion were administered to females. Those tested in period 2 were slightly older than those tested in period 1. In both periods, the majority of tests were performed in the general outpatient department. The most common test strategy in both periods was provider-initiated testing (meaning at the discretion of the clinician), followed by ANC/PMTCT and as part of STI evaluation and care.
Table 1:
Study Population
| Period 1 October 2018-August 2019 |
Period 2 December 2019-March 2020 |
|||
|---|---|---|---|---|
| N (%) | % HIV Positive | N (%) | % HIV Positive | |
| Total | 262,230 (100) | 4.1 | 41,840 (100) | 5.0 |
| Female Sex (at birth) | 155,999 (59.5) | 4.0 | 20,366 (48.7) | 5.3 |
| Non-Citizen | 10,147 (3.9) | 11.5 | 2,572 (6.2) | 22.1 |
| Age * | ||||
| 0–19 years | 26,257 (10.0) | 1.8 | 2,808 (6.7) | 2.6 |
| 20–29 years | 110,169 (42.0) | 2.8 | 15,454 (36.9) | 3.3 |
| 30+years | 125,778 (48.0) | 6.1 | 23,575 (56.3) | 6.4 |
| Testing Point | ||||
| Outpatient | ||||
| ANC/PNC/SRH/PMTCT | 5,610 (2.1) | 2.2 | 1,475 (3.5) | 1.6 |
| Accidents/Emergency | 20,762 (7.9) | 5.0 | 5324 (12.7) | 5.5 |
| General | 195,953 (74.7) | 4.3 | 29,387 (70.2) | 5.1 |
| Departments | 20,297 (7.7) | 3.8 | 767 (1.8) | 3.1 |
| Lab Testing | 526 (0.2) | 20.3 | 129 (0.3) | 8.5 |
| Outreach Services | -- -- | -- | 229 (0.6) | 1.8 |
| Specialty Clinics | 3,054 (1.2) | 2.2 | 848 (2.0) | 6.4 |
| VCT | 9,133 (3.5) | 3.5 | 2,149 (5.1) | 5.1 |
| STI Clinic | 878 (0.3) | 5.8 | 291 (0.7) | 3.1 |
| Pharmacy | - | - | 19 (<0.1) | 10.5 |
| Tuberculosis Clinic | 9 (<0.1) | 0 | - - | - |
| Inpatient | ||||
| ANC/PNC/LD/Gyn | 2,575 (1.0) | 3.9 | 454 (1.1) | 4.0 |
| General | 314 (0.1) | 6.1 | 121 (0.3) | 4.1 |
| ICU | 15 (<0.1) | 6.7 | 6 (<0.1) | 16.7 |
| Medical | 1,315 (0.5) | 4.5 | 261 (0.6) | 8.4 |
| Orthopedics | 626 (0.2) | 5.6 | 172 (0.5) | 6.4 |
| Psychiatric | 24 (<0.1) | 4.2 | - - | - |
| Surgical | 1,129 (0.4) | 4.5 | 205 (0.2) | 2.4 |
| Tuberculosis | 10 (<0.1) | 20.0 | 3 (<0.1) | 0 |
| Testing Strategy | ||||
| ANC/PMTCT | 44,271 (16.9) | 2.2 | 4,621 (11.0) | 2.5 |
| Emergency | 12,289 (4.7) | 4.9 | 3,857 (9.2) | 4.8 |
| Index | 16,389 (6.3) | 11.7 | 4,017 (9.7) | 12.7 |
| Inpatient | 5,013 (1.9) | 4.1 | 1,350 (3.2) | 3..9 |
| Other Provider Initiated Testing | 130,954 (49.9) | 3.6 | 18,461 (44.1) | 3.5 |
| Pediatrics (age<14) | 2,262 (0.9) | 0.8 | 34 (0.4) | 1.3 |
| Pediatric Malnutrition | 109 (<0.1) | 1.8 | 17 (<0.1) | 11.8 |
| Presumptive TB | 7,969 (3.0) | 11.3 | 1,423 (3.4) | 14.3 |
| STI | 26,205 (10.0) | 5.0 | 6,052 (14.5) | 4.8 |
| TB | 1,691 (0.6) | 9.1 | 331 (0.8) | 7.9 |
| VCT | 15,071 (5.8) | 2.3 | 1,500 (3.6) | 2.5 |
| Marital Status | ||||
| Cohabitating | -- -- | -- | 5,311 (12.7) | 8.0 |
| Divorced | -- -- | -- | 253 (0.6) | 9.1 |
| Married | -- -- | -- | 5,641 (13.5) | 5.0 |
| Single | -- -- | -- | 30,191 (72.2) | 4.4 |
| Widowed | -- -- | -- | 444 (1.1) | 7.7 |
| Educational Attainment | ||||
| Minor | -- -- | -- | 546 (1.3) | 1.5 |
| Non-Formal | -- -- | -- | 178 (0.4) | 6.2 |
| None | -- -- | -- | 1,705 (4.1) | 5.1 |
| Primary | -- -- | -- | 3,327 (8.0) | 6.5 |
| Secondary | -- -- | -- | 25,504 (61.0) | 5.7 |
| Tertiary | -- -- | -- | 10,580 (25.3) | 2.8 |
| Employment Status | ||||
| Unemployed | -- -- | -- | 16,167 (38.6) | 5.0 |
| Formally Employed | -- -- | -- | 11,783 (28.2) | 3.8 |
| Informally Employed | -- -- | -- | 5,636 (13.5) | 6.4 |
| Self Employed | -- -- | -- | 5,132 (12.3) | 7.7 |
| Student | -- -- | -- | 2,185 (5.2) | 2.4 |
| Minor | -- -- | -- | 648 (1.6) | 2.3 |
| Retired | -- -- | -- | 289 (0.7) | 3.8 |
| Date of Last Test ** | ||||
| <3 months | -- -- | -- | 27 (<0.1) | 0 |
| 3–6 months | -- -- | -- | 7,275 (17.4) | 2.5 |
| 6–12 months | -- -- | -- | 13,122 (31.4) | 3.0 |
| >12 months | -- -- | -- | 17,977 (43.0) | 6.9 |
| Never Tested | -- -- | -- | 2,582 (6.2) | 9.1 |
| Missing | -- -- | -- | 857 (2.1) | 2.3 |
| Risk Factors | ||||
| Condomless Sex since last HIV test | -- -- | -- | 27,235 (32.8) | 3.6 |
| Condomless Sex with HIV+ partner since last HIV test | -- -- | -- | 5,425 (13.0) | 13.2 |
| Sex under influence of alcohol since last HIV test | -- -- | -- | 9,391 (22.5) | 6.0 |
| History of STIs | -- -- | -- | 7,135 (17.1) | 5.8 |
| TB symptoms | -- -- | -- | 5,682 (13.6) | 11.2 |
| STI symptoms | -- -- | -- | 6,547 (15.7) | 5.7 |
| HIV symptoms | -- -- | -- | 1,510 (3.6) | 10.2 |
26 records missing age in Period 1; 3 records missing age in Period 2
857 records missing date of last test
HIV test positivity was higher in period 2 than in period 1 (5.0% vs. 4.1%, p<0.01). The proportion of those positive for HIV by demographic category varied slightly between the two periods; a slightly higher proportion of females and those age ≥40 tested were positive in period 2 compared to period 1, while 22% of non-citizens were positive in period 2 compared to 11% in period 1. There was some variation in HIV positivity by testing point and testing strategy between periods 1 and 2, particular in categories with few persons tested (Table 1). In both period 1 and period 2, positivity in the sectors and testing strategies in which cessation of testing was determined to be unacceptable was higher than that of other sectors (4.6 vs. 4.2 and 6.3% vs 3.8, both p<0.01).
Behavioral and most demographic risk factors were not collected in Period 1. In Period 2, the majority of those tested were single (unmarried) and had completed secondary or tertiary school. A large proportion (38.6%) were unemployed. Over 90% of persons tested reported having ≥1 HIV test previously, though 43% of those tested had not tested in the prior year. Nearly one-third reported condomless sex and 13% reported condomless sex with a partner known to be living with HIV since the last negative HIV test. Almost one-quarter of those tested reported having sex under the influence of alcohol since the last negative HIV test and 17% reported a history of STIs. Relatively few persons had symptoms of TB, STIs, or HIV. Those who had never tested for HIV (9.1%), reported condomless sex with a partner known to be living with HIV since the last negative HIV test (13.2%), and those with TB (11.2%) or HIV (10.2%) symptoms were most likely to have a positive test result.
Characteristics associated with test positivity in multivariate analysis
In the prediction dataset for period 1 (n=131,115), 4 factors were significantly associated with a positive HIV test result and had adjusted odds ratios ≥1.5 on multivariate analysis (Table 2): citizenship of a nation other than Botswana, age ≥30 years, and the Emergency testing strategy. The coefficients associated with these factors contributed risk score weights ranging from 0.44 (for being tested through the Emergency testing strategy) to 0.87 (citizenship other than Botswana).
Table 2.
Characteristics Associated with Test Positivity by Periodł
| Period 1 N=70,022 |
Period 2 N=10,882 |
|||||
|---|---|---|---|---|---|---|
| Adjusted Odds Ratio | Coefficient | HIV Positive (N=2,621) N (%) |
Adjusted Odds Ratio | Coefficient | HIV Positive (N=407) N (%) |
|
| Female Sex (at birth) | 1.26 (1.17–1.37) | 0.23 | 1,460 (55.7) | 1.80 (1.45–2.24) | 0.59 | 208 (51.1) |
| Non-Citizen | 2.39 (2.07–2.75) | 0.87 | 238 (9.0) | 3.94 (2.95–5.26) | 1.37 | 88 (21.6) |
| Age* (Ref=20–29 years of age) |
||||||
| 0–19 years | 0.73 (0.60–0.88) | −0.32 | 119 (4.5) | 0.48 (0.22–1.04) | −0.74 | 14 (3.1) |
| 30+ years | 2.06 (1.89–2.26) | 0.72 | 1,822 (69.5) | 1.98 (1.51–2.60) | 0.67 | 309 (75.9) |
| Testing Point (ref=General Outpatient Department) | ||||||
| Outpatient | ||||||
| Outreach Services | - | - | - | 0.52 (0.07–3.80) | −0.65 | 1 (0.3) |
| Accidents/Emergency | 0.80 (0.67–0.95) | −0.23 | 339 (12.9) | 0.82 (0.52–1.28) | −0.20 | 97 (23.8) |
| Departments | 0.61 (0.52–0.72) | −0.49 | 164 (6.3) | 0.46 (0.17–1.26) | −0.79 | 4 (1.0) |
| Specialty Clinics | 0.43 (0.28–0.66) | −0.84 | 22 (0.8) | 1.39 (0.79–2.45) | 0.33 | 14 (3.4) |
| Pharmacy | - | - | - | 2.56 (0.30–21.92) | 0.95 | 1 (0.3) |
| Inpatient | ||||||
| General | 0.45 (0.14–1.49) | −0.79 | 3 (0.1) | 1.04 (0.13–8.42) | 0.04 | 1 (0.3) |
| ICU | 4.46 (0.52–40.08) | 1.52 | 1 (0.04) | -- | -- | 0 |
| Medical | 0.78 (0.45–1.31) | −0.25 | 18 (0.7) | 2.77 (1.19–4.45) | 1.01 | 10 (2.5) |
| Orthopedics | 1.26 (0.69–2.28) | 0.23 | 14 (0.5) | 1.81 (0.50–6.61) | 0.59 | 3 (0.7) |
| Psychiatric | - | - | 0 | - | - | 0 |
| Surgical | 1.09 (0.65–1.82) | 0.08 | 23 (0.9) | 0.37 (0.05–2.94) | −0.99 | 1 (0.3) |
| Testing Strategy | ||||||
| Emergency | 1.56 (1.30–1.87) | 0.44 | 296 (11.3) | 1.76 (1.12–2.78) | 0.57 | 95 (23.3) |
| Inpatient | 1.07 (0.76–1.50) | 0.07 | 75 (2.9) | 0.73 (0.38–1.39) | −0.32 | 20 (4.9) |
| Marital Status (ref=Single/Minor) | ||||||
| Cohabitating | - | - | - | 1.55 (1.16–2.07) | 0.44 | 70 (17.2) |
| Divorced | - | - | - | 2.76 (1.32–5.77) | 1.00 | 10 (2.5) |
| Married | - | - | - | 0.66 (0.47–0.91) | −0.43 | 57 (2.5) |
| Widowed | - | - | - | 0.88 (0.38–2.07) | −0.14 | 7 (1.7) |
|
Educational Attainment
(ref=Secondary) |
||||||
| Minor/Non-Formal/None | - | - | - | 0.58 (0.34–1.02) | −0.54 | 18 (4.4) |
| Primary | - | - | - | 0.91 (0.63–1.30) | −0.10 | 43 (10.6) |
| Tertiary | - | - | - | 0.59 (0.44–0.79) | −0.52 | 67 (16.5) |
|
Employment Status
(ref=Formally Employed) |
||||||
| Unemployed | - | - | - | 0.93 (0.69–1.25) | −0.07 | 143 (35.1) |
| Informally Employed | - | - | - | 1.13 (0.80–1.60) | 0.13 | 67 (16.5) |
| Self Employed | - | - | - | 1.64 (1.18–2.27) | 0.49 | 85 (20.9) |
| Minor/Student | - | - | - | 1.08 (0.53–2.23) | 0.08 | 16 (3.9) |
| Retired | - | - | - | 1.34 (0.45–3.98) | 0.28 | 4 (1.0) |
|
Date of Last Test** (Ref=3–6 months) |
||||||
| Never Tested | - | - | - | 2.73 (1.62–4.60) | 1.00 | 45 (11.1) |
| >12 months | - | - | - | 1.97 (1.30–2.99) | 0.68 | 253 (62.2) |
| 6–12 months | - | - | - | 0.85 (0.54–1.33) | −0.17 | 76 (18.7) |
| Risk Factors | ||||||
| Condomless Sex | - | - | - | 1.28 (1.0–1.65) | 0.25 | 303 (74.5) |
| Condomless Sex with HIV+ partner | - | - | - | 3.39 (2.51–4.57) | 1.22 | 70 (17.2) |
| Sex under influence of alcohol | - | - | - | 0.93 (0.72–1.20) | −0.07 | 95 (23.3) |
| History of STIs | - | - | - | 1.01 (0.72–1.40) | 0.01 | 49 (12.0) |
| TB symptoms | - | - | - | 1.93 (1.49–2.50) | 0.66 | 93 (22.9) |
| STI symptoms | - | - | - | 1.30 (0.84–2.02) | 0.26 | 28 (6.9) |
| HIV symptoms | - | - | - | 3.56 (2.39–5.30) | 1.27 | 39 (9.6) |
Bold=significant at p<0.05 and OR>=1.5
Testing points with missing values indicate that no persons at these points tested positive and thus no estimates were produced from the logistic regression
26 records missing age in Period 1; 3 records missing age in Period 2
857 persons missing date of last test
In the prediction dataset for period 2 (n=20,920), 12 factors were significantly associated with having a positive HIV test result and had adjusted odds ratios ≥1.5 on multivariate analysis (Table 2): female sex, citizenship of a nation other than Botswana, age ≥30 years , cohabitation, being divorced, reporting self-employment, never having had an HIV test, having an HIV test >12 months ago, being tested in the inpatient medical ward, being tested under the Emergency testing strategy, and having TB or HIV symptoms. The coefficients associated with these factors contributed risk score weights ranging from 0.44 (for a marital status of cohabitation) to 1.37 (citizenship other than Botswana).
Risk score performance: tests performed, test positivity, and proportion of positives missed
Using a risk score cutoff of ≥1.0, in addition to the testing sectors and strategies where testing would continue as decided a priori, on the period 1 validation dataset (n=131,115) would result in 65,020 (50%) fewer tests performed, a test positivity of 5.2% and would capture 60.9% of the 5,593 positive tests in the validation dataset (Figure 1). Applying these criteria using only variables collected from period 1 to the period 2 validation dataset would result in 10,207 (48.8%) fewer tests performed, a test positivity of 7.1% and would capture 752 of the 1,097 (68.6%) tests in the dataset.
Figure 1.
Performance of risk scores for reduction in tests performed, HIV positive test captured, and test positivity
Using the more complex risk score from period 2 developed with the additional data collected during that period allowed us to investigate the trade-offs of potentially restricting testing to different populations. Limiting testing to persons with a risk score of ≥1.0 would have identified 96.2% of the 1,097 persons testing HIV positive, decreased testing by 13% (n=2,665) and results in a test positivity of 5.9%. Use of a risk score cutoff of ≥2.0 would have identified 83.3% of all persons who tested HIV positive, decreased testing volume by 28% (n=5,939) and resulted in a test positivity of 7.3%. This represents comparable reduction in test volume to the simpler risk score derived using period 1 data, with a higher score sensitivity.
Costs averted
The reductions in testing program costs from application of the various risk score cutoffs, varied by the proportion reduction in administrative costs per test averted, are shown in Table 3. The proportion reductions in total program costs ranged from 12.1–52.3%. The largest reductions in program costs were associated with applying a cutoff score of ≥1.0 to the Period 1 and application of the Period 1 criteria to Period 2 data. Costs averted per positive test missed was highest for a cutoff score of 1.0 in the Period 2 data ($1,163), reflecting the relatively lower number of tests averted and few positives missed. Across all cutoffs in Periods 1 and 2, the proportion of program costs averted was substantially less with assumptions of proportionately lower reductions in administrative costs per test averted.
Table 3.
Proportion of total program costs averted by risk score cutoffs by assumed proportional reductions in administrative costs per test
| Period 1 | Period 2 | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Cutoff 1.0 | Period 1 – Cutoff 1.0 | Cutoff 1.0 | Cutoff 1.5 | Cutoff 2.0 | ||||||
| Administrative Cost per test reduction | Proportion total costs averted | Cost averted per positive test missed | Proportion total costs averted | Cost averted per positive test missed | Proportion total costs averted | Cost averted per positive test missed | Proportion total costs averted | Cost averted per positive test missed | Proportion total costs averted | Cost averted per positive test missed |
| 100% | 52.3% | $548 | 45.6% | $546 | 12.1% | $1,163 | 26.9% | $940 | 38.0% | $839 |
| 75% | 40.8% | $428 | 36.3% | $426 | 9.4% | $906 | 21.0% | $732 | 29.6% | $654 |
| 50% | 29.3% | $307 | 26.1% | $306 | 6.7% | $648 | 15.0% | $524 | 21.2% | $469 |
| 25% | 17.8% | $187 | 15.8% | $186 | 4.1% | $391 | 9.1% | $317 | 12.8% | $283 |
Characteristics of positives missed
We evaluated the characteristics of cases that would be missed using risk scores to assess whether specific populations would be disproportionately impacted by limiting testing (Table 4). Use of a risk score for period 1 equally affected different populations defined by sex and age. In period 2, all score cutoffs led to disproportionate numbers of young and single people not being diagnosed. Very low proportions of positives reporting condomless sex with a partner known to be living with HIV, TB symptoms, or STI symptoms would be missed by any of the proposed risk score cutoffs.
Table 4.
Characteristics of positives captured and missed by risk score cutoffs
| Period 1 | Period 2 | ||||||
|---|---|---|---|---|---|---|---|
| Cutoff 1.0 | Period 1 – Cutoff 1.0 | Cutoff 1.0 | Cutoff 1.5 | Cutoff 2.0 | |||
| N | Positives Missed N=2,242 |
N | Positives Missed N=345 |
Positives Missed N=42 |
Positives Missed N=116 |
Positives Missed N=183 |
|
| Total | 5,596 | % | 1088 | % | % | % | % |
| Female sex (at birth) | 63,154 | 40.8 | 562 | 31.0 | 3.8 | 10.1 | 16.7 |
| Male sex (at birth) | 2,442 | 39.0 | 535 | 32.0 | 2.5 | 7.8 | 15.5 |
| Non-Citizen | 2,446 | 7.9 | 249 | 16.9 | 5.2 | 13.4 | 17.9 |
| Age | |||||||
| 0–19 | 221 | 47.8 | 43 | 46.5 | 2.4 | 11.6 | 20.9 |
| 20–29 | 1,508 | 43.1 | 256 | 32.8 | 10.2 | 19.9 | 27.3 |
| 30+ | 2,113 | 38.3 | 798 | 30.2 | 1.7 | 7.9 | 12.7 |
| Marital status | |||||||
| Minor/Single | -- | 691 | 35.5 | 3.2 | 14.0 | 21.1 | |
| Cohabitating | -- | 224 | 20.1 | 0 | 2.2 | 6.7 | |
| Divorced | -- | 7 | 57.1 | 0 | 0 | 0 | |
| Married | -- | 153 | 27.9 | 3.2 | 8.2 | 12.7 | |
| Widowed | -- | 17 | 41.2 | 0 | 5.9 | 11.8 | |
| Educational Attainment | |||||||
| Minor/Non-Formal/None | -- | 56 | 41.1 | 3.6 | 10.7 | 12.5 | |
| Primary | -- | 112 | 31.3 | 0 | 5.4 | 12.5 | |
| Secondary | -- | 739 | 30.0 | 3.8 | 10.4 | 16.9 | |
| Tertiary | -- | 161 | 35.4 | 6.8 | 14.9 | 19.9 | |
| Date of Last HIV test | |||||||
| Never Tested | -- | 132 | 33.3 | 0 | 2.3 | 9.9 | |
| 3–6 months | -- | 107 | 27.1 | 8.4 | 20.6 | 25.2 | |
| 6–12 months | -- | 210 | 38.1 | 12.4 | 24.3 | 31.9 | |
| >12 months | -- | 639 | 29.3 | 1.1 | 6.3 | 11.9 | |
| Risk Factors | |||||||
| Condomless Sex | -- | 815 | 30.4 | 3.2 | 9.9 | 15.2 | |
| Condomless Sex with HIV+ partner | -- | 3771 | 15.9 | 0 | 0.5 | 1.6 | |
| Sex under influence of alcohol | -- | 293 | 28.7 | 4.1 | 9.9 | 14.7 | |
| History of STIs | -- | 205 | 16.1 | 2.4 | 5.4 | 8.3 | |
| TB symptoms | -- | 339 | 23.5 | 0.3 | 2.0 | 3.2 | |
| STI symptoms | -- | 192 | 7.8 | 0.5 | 2.1 | 4.2 | |
| HIV symptoms | -- | 117 | 27.4 | 0 | 0.8 | 2.6 | |
Bold text: significant at the α=0.05 level
Discussion
In Botswana, a focused approach to HIV testing could reduce test volume by 13–40%, increase test positivity, and potentially reducing program costs by up to 42%, though achieving these outcomes would likely require collection of more detailed demographic, clinical, and behavioral risk data than what is currently done in Botswana and much of Sub-Saharan Africa. At the same time, policy makers should realize that adopting a more focused approach to testing – even on guided by substantial risk data - comes at a cost; curtailing testing would result delayed HIV diagnoses in 4–17% of HIV positive patients and would disproportionately impact younger persons.
Previous efforts to develop risk scores were undertaken for purposes of identifying subpopulations that would most benefit from targeted prevention activities or to improve efficiency of recruitment into population-level prevention studies5–10. While we did identify studies that developed risk scores for use in clinical settings, these were mainly limited the use of an automated screening tool to target testing in persons most at risk for HIV in a US or European setting, typically with the goal of increasing testing rates11–13.
Many of the demographic, clinical, and behavioral risk factors we identified as associated with HIV positivity have been identified in prior studies, suggesting that our risk score may be generalizable in other African settings. Risks identified as being associated with HIV in prior African studies include being a migrant worker14, female gender15, older age15, and cohabitation16. Other factors associated with HIV positivity in our analysis, such as symptoms suggestive of active TB, sex with a known HIV positive partner, and the absence of prior HIV testing and longer time since last HIV testing, make intuitive sense and have some support in the medical literature17–19. Testing in the emergency department likely reflects testing in a population of people who were particularly ill and may have been a form of diagnostic testing. Surprisingly, we did not observe an association between STI symptoms and history of STI and HIV positivity. Many studies have identified bacterial STIs as a risk factor for HIV acquisition20, particularly among women21,22, but it may be that simply asking about symptoms of STI and history of STI are nonspecific when done as part of HIV testing and counseling.
Our findings demonstrate the feasibility of collecting and utilizing risk factors to target HIV testing and illustrates that different approaches to risk-based screening may be useful in different settings. The more complex risk score used in period two, which identified 12 distinct factors strongly associated with HIV positivity, would allow for an approach to maximizes case-finding and efficiency while minimizing the number of missed diagnoses. However, even the most simplified score used in period one could substantially reduce testing volume and increase test positivity, though at a substantial cost in missed diagnoses.
Although successful implementation of any of the risk scores we investigated would likely reduce testing volume and costs, whether the cost savings realized would be worthwhile is uncertain. Employing a more stringent criteria for testing, as in our evaluated score cutoff of ≥2.0, would miss up to 17.0% of positives, resulting in long term implications associated with delayed diagnosis, new transmissions, and increased morbidity and mortality. Notably, persons under age 30 comprised the majority of positives that would be missed under any of the proposed risk score cutoffs evaluated. In general, young people report more inconsistent condom use23 and have more sex partners than older persons (particularly younger males)24,25, and delayed diagnoses in young people may have particularly important impacts on HIV transmission26.
The central issue is whether the cost savings are sufficient to justify the missed diagnoses. Our analysis suggests that implementation of a risk score-based screening approach could decrease program costs. However, our results demonstrate that the magnitude of this cost saving is highly dependent on the reduction in program administrative costs, including staffing, that may vary relatively little with testing volume. For example, testing programs with a reduced volume would still require on-site staff to provide HIV screening and testing services during facility operational hours. Other administrative costs integral to program quality and staff development such as on-site mentoring visits, continuous quality improvement initiatives, and site-based trainings would also likely remain inelastic to reductions in testing volume that would be achieved through even relatively stringent testing criteria. Finally, implementation of risk-score based testing would likely be associated with at least some new costs, particularly if they require programming of the digital data collection instruments to automatically produce a score based on the values of demographic, clinical, and risk factor fields entered by screening staff.
Our analysis is subject to several limitations. Data from the testing sites was periodically matched against a database on persons who previously tested only during period 2, therefore, it is possible that some persons previously diagnosed with HIV were included in the positivity, inflating our estimates of positivity and potentially biasing the risk score cutoff performance in an undetermined direction. The test and program costs cited in this analysis are estimates and may not be reflective of future costs associated with HIV testing programs in Botswana. We recognize that the integration of a risk score into data collection tools, particularly digital data collection tools that require programming and maintenance, may not be feasible in all settings. Finally, the risk factors identified in this study may be of limited generalizability to other settings, particularly given Botswana’s unique epidemiological profile of both high HIV prevalence and substantial progress on meeting UNAIDS goals; however, many of the risks included in our score have been associated with HIV positivity in other settings. Moreover, our results demonstrate that one can use data collectable as part of HIV testing to more narrowly target testing to a higher risk population.
We undertook this study in response to a mandate from PEPFAR to decrease HIV testing volume. Our findings suggest that risk-factor based testing prioritization could improve the efficiency of healthcare facility-based HIV testing; Botswana’s testing program could likely decrease testing volume by approximately 30–40% if it is willing to delay the diagnosis of 11–17% of patients. Whether such a trade-off is worthwhile is uncertain, and will depend on operational considerations, the true impact of reductions in testing volume on program cost, the implications of delayed HIV diagnoses, and the value of alternative uses of funds saved through decreased testing.
Acknowledgments
I-TECH Botswana Testing and Counseling Staff
Conflicts of Interest and Source of Funding:
MG has received support from Hologic. For the remaining authors, none were declared.
This work was funded by President’s Emergency Plan for AIDS Relief through the Health Resources and Services Administration of U.S. Department of Health and Human Services (Cooperative Agreement U91HA06801) and by a grant from the University of Washington / Fred Hutch Center for AIDS Research, an NIH funded program under award number AI027757 which is supported by the following NIH Institutes and Centers: NIAID, NCI, NIMH, NIDA, NICHD, NHLBI, NIA, NIGMS, NIDDK.
References
- 1.UNAIDS, Joint United Nations, Programme on HIV/AIDS. UNAIDS Data 2020. Geneva, Switzerland; 2020. https://www.unaids.org/sites/default/files/media_asset/2020_aids-data-book_en.pdf.AccessedDecember 21, 2020. [Google Scholar]
- 2.PEPFAR 2021 Country and Regional Operational Plan (COP/ROP) Guidance for All PEPFAR Countries.
- 3.Powers KA, Miller WC, Pilcher CD, et al. Improved detection of acute HIV-1 infection in sub-Saharan Africa: Development of a risk score algorithm. AIDS. 2007;21(16):2237–2242. doi: 10.1097/QAD.0b013e3282f08b4d [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Sanders EJ, Wahome E, Powers KA, et al. Targeted screening of at-risk adults for acute HIV-1 infection in sub-Saharan Africa. AIDS. 2015;29(0 3):S221–S230. doi: 10.1097/QAD.0000000000000924 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Menza TW, Hughes JP, Celum CL, Golden MR. Prediction of HIV acquisition among men who have sex with men. Sex Transm Dis. 2009;36(9):547–555. doi: 10.1097/OLQ.0b013e3181a9cc41 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Balkus JE, Brown E, Palanee T, et al. An Empiric HIV Risk Scoring Tool to Predict HIV-1 Acquisition in African Women. J Acquir Immune Defic Syndr. 2016;72(3):333–343. doi: 10.1097/QAI.0000000000000974 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Giovenco D, Pettifor A, MacPhail C, et al. Assessing risk for HIV infection among adolescent girls in South Africa: an evaluation of the VOICE risk score (HPTN 068). J Int AIDS Soc. 2019;22(7). doi: 10.1002/jia2.25359 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Smith DK, Pals SL, Herbst JH, Shinde S, Carey JW. Development of a clinical screening index predictive of incident HIV infection among men who have sex with men in the United States. J Acquir Immune Defic Syndr. 2012;60(4):421–427. doi: 10.1097/QAI.0b013e318256b2f6 [DOI] [PubMed] [Google Scholar]
- 9.Pintye J, Drake AL, Kinuthia J, et al. A Risk Assessment Tool for Identifying Pregnant and Postpartum Women Who May Benefit From Preexposure Prophylaxis. Clin Infect Dis. 2017;64(6):751–758. doi: 10.1093/cid/ciw850 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Kahle EM, Hughes JP, Lingappa JR, et al. An empiric risk scoring tool for identifying high-risk heterosexual HIV-1-Serodiscordant couples for targeted HIV-1 prevention. J Acquir Immune Defic Syndr. 2013;62(3):339–347. doi: 10.1097/QAI.0b013e31827e622d [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Gillet C, Darling KEA, Senn N, Cavassini M, Hugli O. Targeted versus non-targeted HIV testing offered via electronic questionnaire in a Swiss emergency department: A randomized controlled study. PLoS One. 2018;13(3). doi: 10.1371/journal.pone.0190767 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Marcelin JR, Tan EM, Marcelin A, et al. Assessment and improvement of HIV screening rates in a Midwest primary care practice using an electronic clinical decision support system: a quality improvement study. BMC Med Inform Decis Mak. 2016;16(1):76. doi: 10.1186/s12911-016-0320-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Leblanc J, Rousseau A, Hejblum G, et al. The impact of nurse-driven targeted HIV screening in 8 emergency departments: Study protocol for the DICI-VIH cluster-randomized two-period crossover trial. BMC Infect Dis. 2016;16(1). doi: 10.1186/s12879-016-1377-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Marukutira T, Block L, Alwano MG, et al. Comparison of knowledge of HIV status and treatment coverage between non-citizens and citizens: Botswana Combination Prevention Project (BCPP). PLoS One. 2019;14(8). doi: 10.1371/journal.pone.0221629 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Novitsky V, Bussmann H, Okui L, et al. Estimated age and gender profile of individuals missed by a home-based HIV testing and counselling campaign in a Botswana community. J Int AIDS Soc. 2015;18(1). doi: 10.7448/IAS.18.1.19918 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Shisana O, Risher K, Celentano DD, et al. Does marital status matter in an HIV hyperendemic country? Findings from the 2012 South African National HIV Prevalence, Incidence and Behaviour Survey. AIDS Care - Psychol Socio-Medical Asp AIDS/HIV. 2016;28(2):234–241. doi: 10.1080/09540121.2015.1080790 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Corbett EL, Marston B, Churchyard GJ, De Cock KM. Tuberculosis in sub-Saharan Africa: Opportunities, challenges, and change in the era of antiretroviral treatment. Lancet. 2006;367(9514):926–937. doi: 10.1016/S0140-6736(06)68383-9 [DOI] [PubMed] [Google Scholar]
- 18.Vu BN, Green KE, Phan HTT, et al. Lay provider HIV testing: A promising strategy to reach the undiagnosed key populations in Vietnam. PLoS One. 2018;13(12). doi: 10.1371/journal.pone.0210063 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Regan S, Losina E, Chetty S, et al. Factors Associated with Self-Reported Repeat HIV Testing after a Negative Result in Durban, South Africa. PLoS One. 2013;8(4). doi: 10.1371/journal.pone.0062362 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Leon N, Naidoo P, Mathews C, Lewin S, Lombard C. The impact of provider-initiated (opt-out) HIV testing and counseling of patients with sexually transmitted infection in Cape Town, South Africa: A controlled trial. Implement Sci. 2010;5(1):8. doi: 10.1186/1748-5908-5-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Laga M, Manoka A, Kivuvu M, et al. Non-ulcerative sexually transmitted diseases as risk factors for HIV-1 transmission in women: Results from a cohort study. AIDS. 1993;7(1):95–102. doi: 10.1097/00002030-199301000-00015 [DOI] [PubMed] [Google Scholar]
- 22.Wand H, Ramjee G. Biological impact of recurrent sexually transmitted infections on HIV seroconversion among women in South Africa: Results from frailty models. J Int AIDS Soc. 2015;18(1). doi: 10.7448/IAS.18.1.19866 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Keetile M High-risk behaviors among adult men and women in Botswana: Implications for HIV/AIDS prevention efforts. Sahara J. 2014;11(1):158–166. doi: 10.1080/17290376.2014.960948 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Wellings K, Collumbien M, Slaymaker E, et al. Sexual behaviour in context: a global perspective. Lancet. 2006;368(9548):1706–1728. doi: 10.1016/S0140-6736(06)69479-8 [DOI] [PubMed] [Google Scholar]
- 25.Todd J, Cremin I, McGrath N, et al. Reported number of sexual partners: Comparison of data from four African longitudinal studies. Sex Transm Infect. 2009;85(SUPPL. 1):i72. doi: 10.1136/sti.2008.033985 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Lahuerta M, Ue F, Hoffman S, et al. The problem of late ART initiation in sub-Saharan Africa: A Transient aspect of scale-up or a long-term phenomenon? J Health Care Poor Underserved. 2013;24(1):359–383. doi: 10.1353/hpu.2013.0014 [DOI] [PMC free article] [PubMed] [Google Scholar]