Validation of HIV Risk Screening Tool to Identify Infected Adults and Adolescents >14 years at Community Level

Background There are a number of risk factors being used to identify undiagnosed HIV infected adults. As the number of undiagnosed people gets lesser and lesser, it is important to know if existing risk factors and risk assessment tools are valid for use. In this study, we validate existing HIV risk assessment tools and see if they are worth using for HIV case nding among adults who remain undiagnosed. Methods The Tanzania and Zambia Population-Based HIV Impact Assessment (PHIA) household surveys were conducted during 2016. We used adult interview and HIV datasets to assess validity of different HIV risk assessment tools. We rst included 12 risk factors (being divorced, separated or widowed (DSW); having an HIV+ spouse; having one of the following within 12 months of the survey: paid work, slept away from home for at least a month, had multiple sexual partners, paid for sex, had sexually transmitted infection (STI), being a tuberculosis (TB) suspect, being very sick for at least 3 months; had ever sold sex; diagnosed with cervical cancer; and had TB disease into a risk assessment tool and assessed its validity by comparing it against HIV test result. Sensitivity, specicity and predictive value of the tool were assessed against the HIV test result. A receiver operator characteristic (ROC) analysis was conducted to determine a suitable cut-off score in order to have a tool with better sensitivity, specicity, and PPV. ROC comparison statistics was used to statistically test equality between AUC (area under the curve) of the different scores. ROC comparison statistics was also used to determine which risk assessment tool was better compared to the tool that contained all risk factors. Results (95% condence (CI): one present and 8.0% when present. PPV, and (78.6%-85.9%), 41.9% (41.1%-42.7%), 3.2% (2.8%-3.6%), and 99.0% (98.8%-99.3%), respectively. The use of a tool containing conventional risk factors (all except those related with working and sleeping away) was found to have higher AUC compared to the use of all risk factors (p value <0.001), with corresponding sensitivity, specicity, PPV, and NPV of 63.5% (58.9%-68.1%), 66.2% (65.5%-67.0%), 4.2% (3.6%-4.8%), and 98.7% (98.5%-98.9%), respectively. Conclusion Use of a screening tool containing conventional risk factors improved HIV testing yield compared to doing universal testing. Prioritizing people who full multiple risk factors should be explored further to improve HIV testing yield.


Introduction
HIV testing is the gate way for case nding, care and treatment as well as prevention services for high risk individuals. [1][2][3] Over the years remarkable progress has been made to diagnose infected people and put them on treatment. To date, Eastern and Southern African countries have coverage of 87% (77%-95%) for the rst 90 while the coverage is 68% (54%-87%) for Central and Western African countries. [4] This correlates with high uptake of HIV testing across these countries. Prior HIV testing among surveyed men and women 15-49 years was 62% and 74% for Eastern, Southern and Central African countries, respectively from 2015-2018. It was much lower for Western African countries at 31% for women and 16% for men. [5,6] Maintaining such high testing coverage or conducting door-to-door testing in high risk communities is not feasible because of the limitation of funding available for HIV programs considering attening of global support for HIV programs especially that of PEPFAR over the past 10 years. [7,8] As a result of that, a strategic shift has been made to implement targeted HIV testing with the aim of getting high testing yield per dollar spent on HIV test kit in many country HIV programs including high burden countries with the aim of putting as many infected people on treatment and reducing new infection and mortality in the process. [9] A number of HIV risk factors have been identi ed and in use to effect targeted HIV testing of at risk people.
The World Health Organization (WHO) recommends HIV testing for clients having sexually transmitted infection (STI), viral hepatitis, tuberculosis (TB); key populations including commercial sex workers, men having sex with men, and IV drug users; clients with symptoms or medical conditions that could indicate HIV infection, including presumed and con rmed TB cases. [1,2] Other risk factors known to increase risk of HIV infection include having multiple sexual partners [10,11], being divorced, separated or widowed (DSW) [12], history of being a client of a sex worker [13], having cervical Ca [14], being partners with known infected person [15].
A number of HIV risk assessment tools were validated in different settings in an effort to determine best options to identify HIV infected adults. [10,16,17] These tools often don't include risk factors recommended by the WHO and in use in high prevalence countries. Knowing the performance and limitation of a risk screening tool containing all common HIV risk factors is crucial to determine case nding strategies that better t routine implementation setting and assess quality of testing services both in clinical and community settings. This study aims to: HIV testing was offered for everyone in the survey and performed for all consenting adults and adolescents > 14 years during the survey. Known HIV + status was further con rmed through the use of anti-retroviral markers within the blood. Those with anti-retroviral markers were excluded from the study.

Data analysis
Data was obtained from the public domain of PHIA website [22] and analyzed using Stata 14.0 statistical software. First, risk factors that had association with HIV infection among those who never tested for HIV were identi ed using Chi Square test. To develop scores for a risk assessment tool, appropriate screening items were selected and coded one when the risk factor was present and zero when it was not and the total score calculated for each individual as the sum of the numerical values of the screening items included within a tool. For instance, for the rst screening tool where all risk factors were included, the minimum score was 0 while the potential maximum was 12. Chi Square test was also done to examine if having risk screening score of ≥ 1, ≥2, ≥ 3, or ≥ 4 was associated with HIV infection. Sampling weights were used to adjust statistical values taking into account complex sampling design used in PHIA surveys. [20] To determine the optimal cut-off for the screening tool that will enable identi cation of people at risk of HIV infection, a receiver operating characteristic curve (ROC) was plotted. The area under the ROC (Receiver Operating Characteristic) curve (AUC) and corresponding sensitivity, speci city, positive predictive (PPV) and negative predictive values (NPV) by using the screening tool at different level of scores were determined. ROC comparison statistics was used to statistically test equality between AUC of the different scores. For the score selected to be having the best combination of sensitivity, speci city, PPV, NPV and AUC, similar analysis was conducted to see if age, gender, and residence affected AUC. This was done by doing strati ed analysis of AUC using the stated variables.
Finally, to compare and select between the different risk assessment tools, test of quality of AUC was done. Number needed to test to identify one HIV infected person (NNT+) was also calculated to see if risk assessment tools reduced this number compared to universal testing. To select appropriate variables for the second tool, purposeful selection of variables was done using during regression model building. Those with p-value < 0.20 during bi-variable analysis were included in the nal model and examined. Level of signi cance was set at 0.05.
Ethics statement: all surveys had written informed consent, both for interview and blood collection for all participating adults. Parents consented for their children.  Figure 2 summarizes HIV prevalence by risk factor. The highest was recorded for people who sold sex (13.5%), followed by spouses of HIV infected adults (11%) and those who were divorced, separated, or widowed (6.1%). The presence of other risk factors had HIV testing yield ranging from 3.1%-5.5%. TB in the past 10 years had a testing yield of 33.3% for Zambia compared to 3.7% for those without TB in the past 10 years, p value < 0.001 (data not shown).  Fig. 3 indicates the ROC curve comparing the different cut-off points for Tool 1. Area under the curve (AUC) can be seen to reduce as the risk assessment cut-off increases. A score of ≥ 1 was found to have the highest sensitivity at 82.3% (95% CI: 78.6%-85.9%) with the next score of ≥ 2 having nearly half the sensitivity at 46.8% (42.0%-51.6%).
The speci city was higher for a higher cut-off. Positive predictive value was higher for a higher cut-off point while negative predictive value was comparable between all cut-0ff scores. Compared to a cut-off score of ≥ 1, AUC was comparable with a cut-off score of ≥ 2 while it was lower for those with higher cut-off scores (p value < 0.001). (Table 3) The AUC was comparable by age, gender, and residence. (Table 4)    (Table 5) Sensitivity was better for Tool 1 but the corresponding speci city was the lowest. AUC was better for all other tools as compared to this tool and the difference was much higher for Tools 3 and 4 (pvalue < 0.001). (Table 6) PPV or HIV testing yield was highest for Tools 3 and 4 at 4.2% and 4.0%, respectively, if at least one risk factor was present. Tool 3 has the lowest proportion of people eligible for testing at 34%the highest being for tool 1 at 59%. (Table 5) Number needed to test (NNT+) was 24 for Tool

Discussion
We set out to validate HIV risk assessment tool used for adults. In that process we tried various combinations of risk factors in different tools for best possible outcome. The nal tool we recommend for use contains conventional risk factors. That screening tool showed a moderate sensitivity and speci city for identifying infected adults at household level. Using this screening tool, the number needed to test to diagnose one HIV infected adult was 24 down from 43 if universal testing was used.
Looking at individual risk factors, the prevalence of HIV in those who never tested for HIV remained to be high compared to those without risk factors except for TB related risk factors and chronic illness. Two risk factors that stood out with having testing yield of > 10% were selling sex for money and having an HIV + spouse. This is comparable to reported prevalence of 12-20% among FSWs in the study countries. [23] ICT for spouses records even higher testing yield at 32% in program settings. [24] Marital status is an important risk factor. Being divorced widowed or separated was found to be the third highest risk factor with a yield of 6.1%. DSWs are easily identi able at community level and can be used to identify at risk people at community or facility level. It is already a risk factor in many countries. [25,26] Cervical cancer is an important risk factor since Human Papiloma Virus, which is a sexually transmitted viral infection, is the causative agent. [27] Co-infection with HIV was 5.5% in this study. Having multiple sexual partners was found to have a relatively lower prevalence at 3.1%. This may be due to the higher condom use during casual sex with a non-regular partner. [28,29] Lifetime TB disease was not statistically signi cant at the 0.05 cut-off. This should not be misinterpreted as TB not being a risk factor. Data on year of TB diagnosis was present only for Zambia and when we did analysis comparing TB diagnosed in the past 10 years to those who never had TB, or who had TB before 10 years, TB prevalence was much higher at 33.3% prevalence. That should be used in practice instead of lifetime TB disease.
Presumptive TB was not predictor of HIV infection. That is probably because it was de ned broadly especially for cough. A de nition of cough > 2 weeks may make improve the positivity. In studies where the later de nition was used, the positivity was found to be higher. [30,31] Adults having multiple risk factors were found to have high testing yield and were a small fraction of the total assessed. This should be further explored further to identify additional risk factors. A very good example in current use by different case nding and prevention programs is being long distance truck driver, who are likely to sleep away from home, and have multiple sexual partners including sex workers. [32,33] This study also provides some form of reference for the percentage of people who are potentially eligible for HIV testing ful lling at least one of the conventional risk factors among those who never tested for HIV.
In this study, 34.4% adults who never tested for HIV would be eligible for testing. That is around 9.1% of the initial number of adults interviewed. This provides a reference value with which to compare community HIV case nding interventions when such risk factors are used. However, it would not be advisable to test this much adults as it wouldn't be cost effective. A more targeted approach focusing on sex workers and their clients, partners of known HIV + index cases, DSWs, and TB cases would be important starting points. [2] Eliciting some of the risk factors especially those related to sexual history may need some experience especially when implementing the risk assessment tool at community level. The use of health extension workers or community health workers who formally do health interventions may help. At facility level where these risk factors are often used maintaining quality of counselling needs to be ensured through ongoing training and on job coaching. Some of the risk factors are treated in speciality clinics like TB in TB clinic, or STI and cervical cancer in gynaecology clinics for women. This will make it easier to implement universal testing for these groups by providing integrated testing services.
The large number of study participants was one of the strengths of the study. Missing data was minimal and was not related to the risk factors being studied. The performance of the nal tool was found to be independent of age, gender, and residence making the use of the tool applicable in different scenarios. Some of the risk factors were captured a little different from what is used in actual settings. All parameters of screening tool are likely to improve if the presence of the following risk factors was determined for the past 10 years just like what we did with TB instead of just the past 12 months: multiple sexual partners, STI, and paid for sex.