AIDS and Behavior

https://doi.org/10.1007/s10461-020-02962-7

ORIGINAL PAPER

Development and Validation of a Sociodemographic and Behavioral

Characteristics‑Based Risk‑Score Algorithm for Targeting HIV Testing
Among Adults in Kenya
Hellen Muttai1   · Bernard Guyah2 · Paul Musingila1 · Thomas Achia1 · Fredrick Miruka1 · Stella Wanjohi3 ·
Caroline Dande4 · Polycarp Musee5 · Fillet Lugalia6 · Dickens Onyango7 · Eunice Kinywa7 · Gordon Okomo8 ·
Iscah Moth8 · Samuel Omondi9 · Caren Ayieko9 · Lucy Nganga1 · Rachael H. Joseph1 · Emily Zielinski‑Gutierrez1

© The Author(s) 2020

Abstract
To inform targeted HIV testing, we developed and externally validated a risk-score algorithm that incorporated behavioral
characteristics. Outpatient data from five health facilities in western Kenya, comprising 19,458 adults ≥ 15 years tested for
HIV from September 2017 to May 2018, were included in univariable and multivariable analyses used for algorithm develop-
ment. Data for 11,330 adults attending one high-volume facility were used for validation. Using the final algorithm, patients
were grouped into four risk-score categories: ≤ 9, 10–15, 16–29 and ≥ 30, with increasing HIV prevalence of 0.6% [95%
confidence interval (CI) 0.46–0.75], 1.35% (95% CI 0.85–1.84), 2.65% (95% CI 1.8–3.51), and 15.15% (95% CI 9.03–21.27),
respectively. The algorithm’s discrimination performance was modest, with an area under the receiver-operating-curve of 0.69
(95% CI 0.53–0.84). In settings where universal testing is not feasible, a risk-score algorithm can identify sub-populations
with higher HIV-risk to be prioritized for HIV testing.

Keywords  HIV testing · Risk-score algorithm · Kenya

Introduction

With the global commitment to end the human immuno-

Electronic supplementary material  The online version of this
deficiency virus (HIV) epidemic by 2030 [1, 2], the Joint
article (https​://doi.org/10.1007/s1046​1-020-02962​-7) contains United Nations Programme on HIV/AIDS (UNAIDS),
supplementary material, which is available to authorized users. in 2014, set ambitious “90–90–90” targets for accelerat-
ing antiretroviral therapy (ART) coverage [1] in order to
* Hellen Muttai
reduce HIV transmission [3–6] and HIV-related morbidity
[email protected]
and mortality [6] worldwide. The UNAIDS 90–90–90 tar-
1
Division of Global HIV & TB (DGHT), United States gets aim for 90% of people living with HIV to know their
Centers for Disease Control and Prevention (CDC), Kenya, HIV status, 90% of people with diagnosed HIV infection
KEMRI Campus, P.O. Box 606, Nairobi 00621, Kenya
to receive ART, and 90% of people receiving ART to
2
School of Public Health, Maseno University, Kisumu, Kenya achieve viral suppression [1, 2]. In 2018, the Sub-Saharan
3
Center for Health Solution, Kisumu, Kenya Africa region had an estimated 25.6 million people living
4
University of California at San Francisco, Kisumu, Kenya with HIV (68% of all people living with HIV globally)
5 and 1.08 million new HIV infections. By 2018, the region
Elizabeth Glaser Pediatric AIDS Foundation, Homa Bay,
Kenya had achieved 64% ART coverage, with 16.4 million peo-
6 ple accessing ART [7]. Increasing ART coverage in the
Columbia University, Kisumu, Kenya
7
Sub-Saharan Africa region is a global priority. Kenya has
Kisumu County Department of Health, Kisumu, Kenya an adult HIV prevalence of 4.7%, and in 2018, had an
8
Homa Bay County Department of Health, Homa Bay, Kenya estimated 1.6 million people living with HIV and 46,000
9
Siaya County Department of Health, Siaya, Kenya new HIV infections [8]. By September 2018, Kenya had

13
Vol.:(0123456789)
 AIDS and Behavior

achieved 71% ART coverage with over 1.14 million peo- Methods
ple accessing ART [9].
Identifying HIV-positive individuals through HIV test- Study Design
ing services is the single most important step to increas-
ing ART coverage. In line with the 2015 World Health Using a retrospective study design, routinely collected
Organization HIV testing guidelines for generalized HIV HIV testing data from five health facilities in the western
epidemics [10], the 2015 Kenya HIV testing guidelines region of Kenya were used to develop a sociodemographic
recommend provider-initiated testing and counseling for and behavioral characteristics-based risk-score algorithm
all patients attending health facilities [11]. In the last dec- for targeting HIV testing. Data from one high-volume
ade, Sub-Saharan Africa countries, including Kenya, have facility were used to externally validate the algorithm.
made dramatic progress in increasing the coverage of HIV Development and validation of the risk-score algorithm
testing [12–16], leading to fewer undiagnosed people liv- followed systematic methodology that has been well
ing with HIV. Consequently, the percent yield of HIV described [46–49].
diagnoses from universal provider-initiated testing and
counseling in outpatient settings has decreased over time.
In Kenya, 6,103,757 outpatients were tested for HIV in Study Sites and Setting
2019 accounting for 60% of all HIV tests conducted, and
64,493 (1.06%) had an HIV-positive result [9]. Evidence- Homa Bay, Siaya and Kisumu Counties have the highest
based strategies to better target HIV testing could increase HIV prevalence (range 16%–21%) in Kenya, and com-
the HIV-positive yield and improve testing efficiency. bined, have approximately 384,000 people living with HIV
Screening algorithms for HIV testing based on clini- [50]. These counties accounted for 23% of all outpatients
cal characteristics have been evaluated among children tested for HIV nationally in 2017, with 1,696,836 adults
and adolescents [17–21]; however, their use is limited as tested and 23,805 HIV-positive patients identified (1.4%
current guidelines recommend early HIV diagnosis and HIV-positive yield) [51]. Data from seven health facilities
immediate initiation of ART regardless of clinical status that had the highest (1000–5000) average monthly out-
[6, 22–24]. Multiple studies in the United States have patient department visits in the three counties were con-
evaluated behavior-based risk-score algorithms to better sidered for inclusion in our analysis. Although all seven
target routine HIV testing [25–27]. Comparable studies of the selected health facilities routinely collected HIV
in Sub-Saharan Africa are limited. One study conducted behavioral risk data as part of HIV testing and counseling
in Malawi evaluated the use of a risk-score algorithm to services, one facility was found to inconsistently document
identify acute (pre-seroconversion) HIV infection among behavioral risk information and was therefore excluded.
sexually transmitted infection (STI) clinic attendees [28]. The six health facilities included in our study offered
Despite the paucity of studies evaluating risk-score algo- provider-initiated HIV testing and counseling to outpa-
rithms for routine HIV testing in Sub-Saharan Africa, tients using an opt-out approach. This included screening
many studies conducted in this region have documented for HIV-testing eligibility, and provision of pre-test coun-
the association of certain behaviors with higher risk of seling, testing, and post-test counseling to eligible clients.
HIV infection, including: polygamous marriage [29, 30], Eligibility for HIV testing was based on the 2015 Kenya
widowed [31–34] or separated/divorced status [32–34]; Ministry of Health HIV testing guidelines [11], which rec-
having a higher number of sexual partners [31, 35], sex ommend testing individuals who have never been tested
in exchange for money or other favors [36, 37], or casual for HIV; individuals whose last reported negative HIV test
heterosexual sex [38]; being a sex worker, or man who result was more than 12 months ago, or who do not know
has sex with men [38]; injection drug use [38]; fish trade the date of their most recent HIV test; individuals who
[38]; inconsistent condom use [35]; use of alcohol before have signs, symptoms, or a diagnosis of tuberculosis or
sex [39, 40]; intimate partner violence [41]; having an STI; and those who report recent HIV exposure. In March
HIV infected sexual partner [42]; having an STI [31, 35]; 2017, eligibility for HIV testing was expanded in order to
and uncircumcised status in men [43–45]. increase access to HIV testing services. The expanded eli-
To inform targeted HIV testing, we developed and gibility criteria included individuals reporting a negative
validated a risk-score algorithm that incorporates sexual HIV test result in the past 3 to 12 months, and those report-
behavioral characteristics and assessed its performance ing a negative HIV test result in the past < 3 months, but
among adults attending routine outpatient services at for whom the test result could not be confirmed in clinic
selected health facilities in Kenya. records. Eligible patients were tested for HIV according to
the Ministry of Health guidelines using Determine™ and

13
AIDS and Behavior

First Response™ rapid point-of-care kits; an individual to be associated with HIV infection. These included: soci-
was considered HIV-negative (uninfected) if the Deter- odemographic characteristics (sex, age, marital status and
mine test result was negative, HIV-positive (infected) if occupation); behavioral characteristics (change in sexual
the Determine and First Response serial test results were partners, number of sexual partners, consistent condom
positive, and inconclusive if the Determine result was use, had sex in exchange for money/favors, engaged in sex
positive and the First Response result was negative. From work, men who reported having sex with men, female anal
September 2017, the health facilities in our study used sex, injecting drugs for pleasure, had sex under the influ-
standardized forms to document behavioral risk charac- ence of alcohol or other substance, and coerced to have
teristics routinely assessed by HIV-testing counselors to sex); reported treatment for STI; circumcision status; and
guide HIV prevention counseling during pre-test coun- specific reasons for HIV testing eligibility (never tested for
seling sessions. HIV, interval since last HIV-negative test, having tubercu-
Our analysis included data from clients aged 15 years and losis, having an STI, and reporting recent HIV exposure).
older who were tested for HIV between September 2017 and Characteristics such as education level, having an HIV
May 2018 in the outpatient departments of the 6 study sites, infected sexual partner [52] and involvement in fish trade
and who had documentation of one or more behavioral risk [38], which have been shown to be associated with HIV
characteristics. Records for patients with inconclusive HIV infection in other studies, were not routinely collected.
test results were excluded. At the six health facilities, data Development of the HIV infection predictive model
for an entire month were excluded if ≥ 50% of patients tested was conducted in a systematic fashion, using univari-
for HIV in that month did not have any documentation of able and multivariable analyses. As recommended for
behavioral risk characteristics. continuous variables [55–57], the association between
age and HIV infection were assessed using a generalized
Data Management additive model; the predicted odds of HIV-positivity by
age were plotted, and informed age categorization into
Sociodemographic, HIV screening and testing, and behavio- 5-year bands. The age-bands were further categorized into
ral risk information were recorded manually on Ministry of groups according to their HIV prevalence (the proportion
Health registers and standardized forms. At each health facil- of HIV infected individuals) as follows: ages 15–19, 20–24
ity, the data were reviewed for completeness and accuracy, and ≥ 50 years (HIV prevalence range of 0.33%–0.99%);
and entered into a secure password-protected database with ages 25–29, 30–34 and 45–49 years (HIV prevalence range
in-built data consistency checks. Data meeting the study of 1.32%–1.68%); and ages 35–39 and 40–44 years (HIV
inclusion criteria were stripped of all identifiers (names and prevalence range of 1.97%–2.49%).
unique patient numbers), assigned new evaluation-specific Univariable analysis was conducted to assess the inde-
identification numbers, entered into a study-specific secure pendent association between the sociodemographic and
password-protected database, and encrypted. Encrypted de- behavioral characteristics and HIV infection, by comput-
identified data were uploaded from each facility to a central ing odds ratios (ORs) and their corresponding 95% confi-
database. dence intervals (CIs) and p values (significant at p ≤ 0.05).
Two variables were not included in the univariable analy-
Risk‑Score Algorithm Development sis: having sex in the prior 12 months, as multiple charac-
teristics were assessed only for those who had sex in the
Adult outpatient HIV testing data from five of the six health prior 12 months, and consistent condom use with a sexual
facilities included in our study were used to develop over- partner, as the documentation format made this variable
all and gender-specific risk-score algorithms; two facilities difficult to interpret.
were in Kisumu County (a referral hospital and sub-county The initial full multivariable analysis included all vari-
hospital), two were in Homa Bay County (a county and sub- ables with a significant higher odds (OR > 1.0) of HIV
county hospital), and one was in Siaya County (a county infection in univariable analysis, and those selected based
hospital). These five facilities accounted for approximately on prior knowledge of an association with HIV infec-
7% of adult outpatients tested for HIV in the three counties tion. The variables in the full multivariable analysis were
in 2017. evaluated in a stepwise multivariable logistic regression,
The primary outcome in this analysis was an HIV-posi- that incorporated Akaike information criterion for model
tive test result. Sociodemographic and behavioral charac- selection, to identify the model/algorithm that best pre-
teristics were considered for inclusion in the development dicted HIV infection. Corresponding ORs, β regression
of the predictive model if they were among those routinely coefficients and 95% CIs were computed. All participants
collected during the pre-test counseling phase of HIV test- with missing data were excluded from the univariable and
ing, and have been shown [27, 33, 52–54] or hypothesized multivariable analyses.

13
 AIDS and Behavior

The final model was internally validated using 10-fold Ethical Considerations
cross-validation. The ability of the final risk-score algo-
rithm to discriminate between individuals with, and with- The Institutional Review Board of Kenyatta National Hos-
out, HIV infection was evaluated by computing the average pital (Nairobi, Kenya) approved the protocol to conduct
area under the receiver operating curve (AUC, the area this analysis. The protocol was also reviewed and approved
under a plot of sensitivity and the inverse of specificity) according to the human research protection procedures for
from the ten different cross-validation models. R-squared the United States Centers for Disease Control and Prevention
­(R2) was computed to assess the extent to which the HIV (Atlanta, Georgia).
prevalence variability can be explained by the model.
Risk-scores for each variable in the final model were
created by multiplying the corresponding β regression Results
coefficient by 10 and rounding to the nearest integer for
ease of calculation. Each patient’s total risk-score was gen- Characteristics of Patients at the Five Health
erated by summing the scores for all variables met. Facilities Used for Risk‑Score Algorithm
To create risk-score categories, patient risk-scores were Development
arranged in ascending order. The corresponding HIV prev-
alence for patients meeting each score was computed and Out of the 45 total months (9 months for each of the 5 health
used to identify mutually exclusive cut-points for unique facilities) that data were eligible for inclusion in the study,
risk-score groupings. The aggregate HIV prevalence and data for 37 (82%) months met the inclusion criteria. Dur-
corresponding CIs were then calculated for each defined ing these months, 99.9% (27,685/27,692) of adults attend-
risk-score grouping. ing OPD services were screened for HIV testing eligibility,
and 87% (21,764/24,966) of those eligible were tested for
HIV. Of 21,745 patients with positive or negative HIV test
Risk‑Score Algorithm Validation results, 19,458 (89%) had behavioral risk characteristics
documented and were included in our analysis.
Data from Kisumu County Hospital, a health facility Among the 19,458 patient records included, the median
among the six high-volume sites selected for inclusion age was 29  years (interquartile range 22–43  years) and
in our study, were used to externally validate the overall 11,149 (57%) were women (Table 1). Most patients [10,731
and the gender-specific risk-score algorithms developed. (61%)] were in monogamous marriage, and approximately
This hospital had the highest number (~ 38,000) of adult two-thirds were either in trade/sales/service occupation
outpatients tested for HIV in 2017 in the three Counties [5467 (29%)] or were school/college going [5167 (27%)].
of Siaya, Kisumu and Homa Bay. Procedures for HIV test- The majority of patients [18,450 (95%)] reported having
ing, documentation of sociodemographic and behavioral sex in the prior 12 months, of whom 5038 (28%) reported
characteristics, and management of HIV testing data were having 2 or more sexual partners, and 2749 (17%) reported
similar to those earlier described for the other facilities changes in sexual partners. Among those with changes in
included in the study. sexual partners, 1411 (51%) reported new sexual partners
For validation, each patient’s risk-score was generated and 800 (29%) were widowed. Few patients reported having
using the risk-score algorithm developed, and patients were sex in exchange for money/favors/gifts [773 (4%)], having
grouped into respective risk-score categories. HIV preva- sex under the influence of alcohol/other substances [496
lence and corresponding CIs for each risk-score category (3%)], having been coerced to have sex [480 (3%)], or hav-
were then calculated. The AUC and ­R2 were computed in ing received treatment for STI in the prior 12 months [251
order to assess the algorithm’s discrimination performance, (1%)]. A minority of patients had never been tested for HIV
and the extent to which variability in HIV prevalence is [688 (3%)] or had a negative HIV test result > 12 months
explained by the model, respectively. prior [12 (0.1%)] (Table 1). Overall, 210 (1.1%) patients
were HIV-positive.
Compared to women, a significantly higher proportion
Data Analysis of men were never married (30% vs 23%, p < 0.001), in a
polygamous marriage (9% vs 3%, p < 0.001), in a manual/
Data were managed using Stata Statistical Software version domestic occupation (12% vs 1%, p < 0.001), had ≥ 2 sexual
14 (StataCorp, College Station, TX) and R version 3.6.2 partners (35% vs 22%, p < 0.001) and reported a new sexual
[58]. The Classification And REgression Training (caret) partner in the prior 12 months (13% vs 6%, p < 0.001). Con-
package for predictive modelling was used to perform versely, a significantly higher proportion of women were in
10-fold cross-validation and to compute the AUC and ­R2. monogamous marriage (64% vs 57%, p < 0.001), widowed

13
AIDS and Behavior

Table 1  Characteristics of outpatient attendees by gender at five high- Sub-county Hospital, Mbita Sub-county Hospital) used for algorithm
volume facilities (Jaramogi Oginga Odinga Teaching and Referral development
Hospital, Homa Bay County Hospital, Siaya County Hospital, Ahero

Characteristic All patients Men Women p value

n (%) n (%) n (%)

Total 19,458 8309 11,149

Sociodemographic characteristics
 Age in years, median (interquartile range) 29 (22–43) 30 (22–43) 29 (22–45)
 Age categories
  15–19, 20–24 and ≥ 50 years 10,577 (54) 4399 (53) 6178 (55) 0.042
  25–29, 30–34 and 45–49 years 6211 (32) 2668 (32) 3543 (32) 1.00
  35–39 and 40–44 years 2670 (14) 1242 (15) 1428 (13) 0.14
 Marital status
  Never married 4546 (26) 2248 (30) 2298 (23)  < 0.001
  Married monogamous 10,731 (61) 4312 (57) 6419 (64)  < 0.001
  Married polygamous 952 (6) 667 (9) 285 (3)  < 0.001
  Cohabiting 239 (1) 59 (1) 180 (2) 0.61
  Separated/divorced 224 (1) 118 (1) 106 (1) 0.54
  Widowed 948 (5) 181 (2) 767 (7) 0.004
 Occupation
  Professional/administrative/clerical 2180 (11) 1011 (12) 1169 (11) 0.15
  Manuala/domestic 1044 (6) 931 (12) 113 (1)  < 0.001
  Agriculture 2285 (12) 1050 (13) 1235 (11) 0.14
  Trade/sales/service 5467 (29) 1926 (24) 3541 (32)  < 0.001
  Unemployed 2924 (15) 863 (11) 2061 (19)  < 0.001
  School/college going 5167 (27) 2282 (28) 2885 (26) 0.11
Behavioral characteristics
 Had sex in the prior 12 months
  Yes 18,450 (95) 7840 (94) 10,610 (95) 0.003
  No 1008 (5) 469 (6) 539 (5) 0.49
 Number of sexual partners in the prior 12 months
  1 13,220 (72) 5084 (65) 8136 (78)  < 0.001
  ≥2 5038 (28) 2681 (35) 2357 (22)  < 0.001
 Changes in sexual partners in the prior 12 months
  Not had change in sexual partner 13,523 (83) 5459 (81) 8064 (85)  < 0.001
  New sexual partner 1411 (8) 867 (13) 544 (6)  < 0.001
  Newly married 155 (1) 69 (1) 86 (1) 1.00
  Ended a sexual relationship 293 (2) 131 (2) 162 (2) 1.00
  Divorced/separated 90 (1) 44 (1) 46 (0.4) 0.49
  Widowed 800 (5) 175 (2) 625 (6) 0.051
 Had sex in exchange for money/favors in the prior 12 ­monthsb n = 17,373 n = 7358 n = 10,015
  Yes 773 (4) 305 (4) 468 (5) 0.52
 Had sex under influence of alcohol/other substance in the prior 12 months n = 17,366 n = 7354 n = 10,012
  Yes 496 (3) 321 (4) 175 (2) 0.23
 Coerced to have sex in the prior 12 months n = 17,094 n = 7274 n = 9820
  Yes 480 (3) 97 (1) 383 (4) 0.15
 Reported treatment for STI in the prior 12 months n = 16,928 n = 7188 n = 9740
  Yes 251 (1) 121 (2) 130 (1) 0.51
 Engaged in sex work, men who have with men, female anal sex, injecting n = 16,450 n = 6960 n = 9490
­ onthsb
drugs for pleasure in the prior 12 m
  Yes 730 (4) 294 (4) 436 (5) 0.53
Circumcision status (men only) n = 6158

13
 AIDS and Behavior

Table 1  (continued)
Characteristic All patients Men Women p value
n (%) n (%) n (%)

 Circumcised 4871 (78)

HIV testing information
 Reason for HIV testing eligibility
  Never been tested for HIV 688 (3) 358 (4) 330 (3) 0.48
  HIV negative test > 12 months prior 12 (0.1) 7 (0.08) 5 (0.05) 0.84
  HIV negative test 6 to 12 months prior 5967 (31) 2651 (32) 3316 (30) 0.10
  HIV negative test 3 to 6 months prior 9454 (49) 3913 (47) 5541 (50) 0.004
  HIV negative test < 3 months ago (unverified) 2666 (14) 1067 (13) 1599 (14) 0.46
  HIV negative test date unknown 668 (3) 311 (4) 357 (3) 0.48
  Has tuberculosis, STI or recent HIV exposure 3 (0.02) 2 (0.02) 1 (0.009) 0.95

n number, STI sexually transmitted infection

a
 Manual occupation refers to both skilled and unskilled
b
 Due to multicollinearity, the characteristic “engaged in sex work, men who have with men, female anal sex, injecting drugs for pleasure in the
prior 12 months” was excluded in univariable and multivariable analysis, while the characteristic “had sex in exchange for money/favors in the
prior 12 months” was included

(7% vs 2%, p 0.004), in a trade/sale/service occupation (32% The variables in the final algorithm were each assigned a
vs 24%, p < 0.001), unemployed (19% vs 11%, p < 0.001), or risk-score, and each patient’s risk-score was calculated as the
reported being widowed in the prior 12 months (6% vs 2%, sum of risk-scores for variables met. Patients were grouped
p 0.051, Table 1). into the following 4 risk-score categories: ≤ 9 [HIV preva-
lence 0.6% (95% CI 0.46–0.75)], 10–15 [HIV prevalence
Overall Risk‑Score Algorithm Development 1.35% (95% CI 0.85–1.84)], 16–29 [HIV prevalence 2.65%
(95% CI 1.8–3.51)], and ≥ 30 [HIV prevalence 15.15%
The following characteristics were positively significantly (95% CI 9.03–21.27)] (Table 4). The 3 highest risk-score
associated with HIV infection in univariable analysis: being categories (score ≥ 10) accounted for 55% of HIV-positive
aged 35–39 and 40–44 years; male gender; manual/domestic patients identified, yet represented just 24% of the total
and trade/sales/service occupation; polygamous marriage, patients tested for HIV. Similarly, patients in the 2 highest
separated/divorced or widowed; in the prior 12 months hav- risk-score categories (score ≥ 16) accounted for 37% of HIV-
ing a new sexual partner, ≥ 2 sexual partners, or reporting positive patients identified, yet represented just 10% of the
treatment for STI; having never been tested for HIV; or hav- total patients tested for HIV.
ing a negative HIV test result > 12 months prior (Table 2).
The initial full multivariable analysis included all the Overall Risk‑Score Algorithm Validation
variables that were positively significantly associated with
HIV infection in the univariable analysis. Additional vari- The validation dataset consisted of 11,330 patient records, of
ables that were also included based on known association which 174 (1.6%) were HIV-positive. The sociodemographic
with HIV infection were: divorced/separated or widowed, and behavioral characteristics of patients in the validation
in the prior 12 months having sex in exchange for money/ dataset are shown in Table 5. In comparison to the devel-
favors and coerced to have sex (Table 3). The AUC for the opment dataset, the validation dataset had a significantly
full model was 0.66 (95% CI 0.44–0.88). higher proportion of patients with manual/domestic and
The final best-fit model/risk-score algorithm consisted of trade/sales/service occupation, and a significantly lower
the following variables: age category 35–39/40–44 years; proportion of patients who reported having ≥ 2 sexual part-
occupation (manual/domestic or trade/sales/service); mari- ners in the prior 12 months, and having a negative HIV test
tal status (polygamous marriage, separated/divorced or wid- result > 12 months prior (Table 5).
owed); in the prior 12 months having ≥ 2 sexual partners or When applied to the validation dataset, the final risk-score
reporting treatment for an STI; and having never been tested algorithm/model had an AUC of 0.69 (95% CI 0.60–0.77)
for HIV or having a negative HIV test result > 12 months and ­R2 of 0.88. The risk score categories ≤ 9, 10–15, 16–29
prior (Table 3). The final model/algorithm had an AUC of and ≥ 30 had an increasing HIV prevalence of 0.97% (95%
0.69 (95% CI 0.53–0.84) and ­R2 of 0.89. CI 0.76–1.18), 2.32% (95% CI 1.47–3.17), 3.69% (95% CI
2.62–4.76) and 6.76% (95% CI 1.04–12.48), respectively

13
AIDS and Behavior

Table 2  Univariable association of sociodemographic and behavioral characteristics with HIV infection at five high-volume facilities used for
algorithm development
Characteristic Number HIV positive/tested Univariable ­analysisa p value
(%)
Odds ratio (95% CI)

Total 210/19,458 (1.08)

Ages 15–19, 20–24 and ≥ 50 years 74/10,577 (0.70) 0.46 (0.32, 0.64)  < 0.001
Ages 25–29, 30–34 and 45–49 years 79/6211 (1.27) 1.2 (0.87, 1.66) 0.28
Ages 35–39 and 40–44 years 57/2670 (2.13) 2.39 (1.68, 3.39)  < 0.001
Men 110/8309 (1.32) 1.46 (1.06, 2.01) 0.019
Women 100/11,149 (0.90) 0.68 (0.5, 0.94) 0.019
Never married 32/4546 (0.70) 0.5 (0.31, 0.83) 0.006
Married monogamous 112/10,731 (1.04) 0.74 (0.53, 1.02) 0.06
Married polygamous 20/952 (2.10) 2.11 (1.25, 3.56) 0.005
Cohabiting 2/239 (0.84) 0.88 (0.22, 3.58) 0.86
Separated/divorced 16/224 (7.14) 7.56 (4.01, 14.25)  < 0.001
Widowed 17/948 (1.79) 2.3 (1.32, 4.02) 0.003
Professional/administrative/clerical occupation 22/2180 (1.01) 0.71 (0.42, 1.22) 0.22
Manual/domestic occupation 22/1044 (2.11) 2.06 (1.26, 3.39) 0.004
Trade/sales/service occupation 93/5467 (1.70) 1.84 (1.33, 2.53)  < 0.001
Agriculture occupation 22/2285 (0.96) 0.82 (0.49, 1.39) 0.47
School/college going 18/5167 (0.35) 0.24 (0.13, 0.45)  < 0.001
Unemployed 28/2924 (0.96) 1.22 (0.77, 1.95) 0.39
≥ 2 Sexual partners in the prior 12 months 83/5038 (1.65) 2.29 (1.62, 3.22)  < 0.001
No change in sexual partners in prior 12 months 126/13,523 (0.93) 0.52 (0.36, 0.77) 0.001
New sexual partner in prior 12 months 27/1411 (1.91) 2.61 (1.69, 4.03)  < 0.001
Newly married in prior 12 months 1/155 (0.65) 0.66 (0.09, 4.78) 0.68
Ended a sexual relationship in prior 12 months 2/293 (0.68) 0 (0, 1.01) 0.97
Divorced/separated in prior 12 months 4/90 (4.44) 3.09 (0.75, 12.76) 0.12
Widowed in prior 12 months 6/800 (0.75) 1.03 (0.45, 2.34) 0.94
Had sex in exchange for money/favors in prior 12 months 13/773 (1.68) 1.58 (0.83, 3.02) 0.16
Had sex under influence of alcohol/other substance in prior 5/496 (1.01) 0.77 (0.24, 2.42) 0.65
12 months
Coerced to have sex in prior 12 months 8/480 (1.67) 1.43 (0.63, 3.26) 0.39
Reported treatment for STI in prior 12 months 9/251 (3.59) 3.34 (1.55, 7.22) 0.002
Never been tested for HIV 32/688 (4.65) 5.44 (3.4, 8.71)  < 0.001
HIV negative test > 12 months prior 1/12 (8.33) 13.89 (1.7, 113.55) 0.014
HIV negative test 6 to 12 months prior 63/5967 (1.06) 1.13 (0.8, 1.58) 0.49
HIV negative test 3 to 6 months prior 74/9454 (0.78) 0.61 (0.44, 0.85) 0.003
HIV negative test < 3 months ago (unverified) 36/2666 (1.35) 0.93 (0.57, 1.53) 0.79
HIV negative test date unknown 4/668 (0.60) 0.58 (0.19, 1.84) 0.36
Has tuberculosis, STI or recent HIV exposure 0/3 (0.00)

n number, CI confidence interval, STI sexually transmitted infection

a
 Missing data omitted from univariable analysis

(Table 4). The 3 highest risk-score categories (score ≥ 10) Development of Gender‑Specific Risk‑Score

accounted for 49% of HIV-positive patients identified, but Algorithms
only 23% of the total patients tested for HIV. The 2 high-
est risk-score categories (score ≥ 16) accounted for 31% of Characteristics that were positively significantly associ-
HIV-positive patients identified, but only 12% of the total ated with HIV infection in univariable analysis (OR > 1.0 at
patients tested for HIV. p ≤ 0.05) among men and women are shown in Supplemen-
tary Table SI. Full multivariable models for men and women

13
 AIDS and Behavior

Table 3  Multivariable association of sociodemographic and behavioral characteristics with HIV infection among outpatient attendees at five
high-volume facilities
Full multivariable model Stepwise multivariable analysis
Odds ratio β (95% CI) Odds ratio β (95% CI) Risk ­scorea

Ages 35–39 and 40–44 years 2.12 0.75 (0.38, 1.12) 2.16 0.77 (0.4, 1.14) 8
Men 1.16 0.15 (− 0.21, 0.52)
Manual/domestic occupation 1.99 0.69 (0.11, 1.26) 2.20 0.79 (0.23, 1.35) 8
Trade/sales/service occupation 1.92 0.65 (0.29, 1) 1.95 0.67 (0.31, 1.02) 7
Married polygamous 1.55 0.44 (− 0.15, 1.04) 1.80 0.59 (0.01, 1.17) 6
Widowed 3.90 1.36 (0.66, 2.06) 2.39 0.87 (0.26, 1.48) 9
Separated/divorced 6.96 1.94 (1.19, 2.68) 5.26 1.66 (0.98, 2.34) 17
≥ 2 Sexual partners in prior 12 months 1.63 0.49 (− 0.04, 1.02) 1.58 0.46 (0.06, 0.86) 5
New sexual partner in prior 12 months 1.27 0.24 (− 0.4, 0.88)
Divorced/separated in prior 12 months 0.33 − 1.1 (− 2.73, 0.52)
Widowed in prior 12 months 0.39  − 0.95 (− 2.03, 0.13)
Coerced to have sex in prior 12 months 1.32 0.28 (− 0.61, 1.17)
Had sex in exchange for money/favors in prior 12 months 0.90  − 0.11 (− 0.85, 0.63)
Reported treatment for STI in prior 12 months 2.61 0.96 (0.13, 1.79) 2.97 1.09 (0.29, 1.9) 11
Never been tested for HIV 6.23 1.83 (1.34, 2.33) 6.17 1.82 (1.33, 2.31) 18
HIV negative result > 12 months ago 7.92 2.07 (− 0.14, 4.29) 9.03 2.2 (0.02, 4.37) 22

β Regression coefficient, CI confidence interval, STI sexually transmitted infection

a
 Computed by multiplying the β regression coefficients by 10 and rounding to the nearest integer

are shown in Supplementary Tables SII and SIII. The AUC (separated/divorced or widowed); in the prior 12 months
Table 4  Final algorithm Risk-score category Number HIV HIV prevalence, % (95% CI) % of total % of total tests
risk-score categories for positive/tested HIV positive
development and validation
datasets Risk-score categories for algorithm development dataset
 ≤ 9 68/11,289 0.6 (0.46, 0.75) 45 76
 10–15 28/2076 1.35 (0.85, 1.84) 18 14
 16–29 36/1357 2.65 (1.8, 3.51) 24 9
 ≥ 30 20/132 15.15 (9.03, 21.27) 13 1
 Totala 152/14,854 1.02%
Risk-score categories for algorithm validation dataset
 ≤ 9 79/8142 0.97 (0.76, 1.18) 51 77
 10–15 28/1207 2.32 (1.47, 3.17) 18 11
 16–29 44/1193 3.69 (2.62, 4.76) 28 11
 ≥ 30 5/74 6.76 (1.04, 12.48) 3 1
 Totala 156/10,616 1.47%

CI confidence interval
a
 Patients with missing data omitted from the analysis

for the full model was 0.75 (95% CI 0.65–0.85) among men having ≥ 2 sexual partners or a new sexual partner; circum-
and 0.68 (95% CI 0.56–0.8) among women. cised status; and having never been tested for HIV (Supple-
The final best-fit model/risk-score algorithm among men mentary Table SII).
had an AUC of 0.76 (95% CI 0.56–0.96) and an R ­ 2 of 0.69, The final risk-score algorithm among women had an AUC
and consisted of the following variables: age categories of 0.66 (95% CI 0.47–0.85) and an ­R2 of 0.87, and consisted
25–29/30–34/45–49 years and 35–39/40–44 years; occupa- of the following variables: age category 35–39/40–44 years;
tion (manual/domestic or trade/sales/service); marital status trade/sales/service occupation; marital status (polygamous

13
AIDS and Behavior

Table 5  Comparison of Characteristic Five health ­facilitiesa One health ­facilityb p value

characteristics of patients at
the five health facilities used n (%) n (%)
for algorithm development and
Total 19,458 11,330
one facility used for algorithm
validation Sociodemographic characteristics
 Age in years, median (interquartile range) 29 (22–43) 27 (22–38)
 Age categories
  15–19, 20–24 and ≥ 50 years 10,577 (54) 5744 (51)  < 0.001
  25–29, 30–34 and 45–49 years 6211 (32) 4089 (36)  < 0.001
  35–39 and 40–44 years 2670 (14) 1497 (13) 0.37
 Gender
  Men 8309 (43) 4706 (42) 0.27
  Women 11,149 (57) 6624 (58) 0.19
 Marital status
  Never married 4546 (26) 3968 (35)  < 0.001
  Married monogamous 10,731 (61) 6191 (55)  < 0.001
  Married polygamous 952 (6) 465 (4) 0.4
  Cohabiting 239 (1) 11 (0.1) 0.74
  Separated/divorced 224 (1) 154 (1) 1
  Widowed 948 (5) 532 (5) 1
 Occupation
  Professional/administrative/clerical 2180 (11) 1086 (10) 0.38
  Manual/domestic 1044 (6) 1000 (9)  < 0.001
  Agriculture 2285 (12) 643 (6)  < 0.001
  Trade/sales/service 5467 (29) 3783 (34)  < 0.001
  Unemployed 2924 (15) 1974 (18) 0.005
  School/college going 5167 (27) 2531 (23)  < 0.001
 ≥ 2 Sexual partners in prior 12 months 5038 (28) 1181 (11)  < 0.001
 Reported treatment for STI in prior 12 months 251 (1) 74 (1) 1
 Reason for HIV testing eligibility
  Never been tested for HIV 688 (4) 695 (6) 0.08
  HIV negative test > 12 months prior 12 (0.1) 2 (0.02)  < 0.001
  HIV negative test 6 to 12 months prior 5967 (31) 3623 (32) 0.31
  HIV negative test 3 to 6 months prior 9454 (49) 4836 (43)  < 0.001
  HIV negative test < 3 months ago (unverified) 2666 (14) 2074 (18)  < 0.001
  HIV negative test date unknown 668 (3) 99 (1) 0.26
  Has tuberculosis, STI or recent   HIV exposure 3 (0.02) 1 (0.01)  < 0.001

n number, STI sexually transmitted infection

a
 Jaramogi Oginga Odinga Teaching and Referral Hospital, Homa Bay County Hospital, Siaya County Hos-
pital, Ahero Sub-county Hospital, Mbita Sub-county Hospital
b
 Kisumu County Hospital

marriage, separated/divorced or widowed); in the prior for HIV. Similarly, among women, the 3 highest risk-score
12 months having a new sexual partner or reporting treat- categories (score ≥ 8) accounted for 51% of HIV-positive
ment for an STI; and having never been tested for HIV or patients identified, yet represented 23% of the total patients
having a negative HIV test result > 12 months prior (Sup- tested for HIV (Supplementary Table SIV).
plementary Table SIII).
Risk-score categories and corresponding HIV preva- Validation of the Gender‑Specific Risk‑Score
lence among men and women are shown in Supplementary Algorithm
Table SIV. Among men, the 3 highest risk-score categories
(score ≥ 13) accounted for 86% of HIV-positive patients The validation dataset comprised 4706 (42%) men and
identified, yet represented 50% of the total patients tested 6624 (58%) women. When applied to the validation dataset,

13
 AIDS and Behavior

the final algorithm/model had an AUC of 0.71 (95% CI development and validation datasets, respectively). Perfor-
0.57–0.86) and an ­R2 of 0.85 among men, and an AUC of mance of the algorithm in discriminating patients with, and
0.66 (95% CI 0.49–0.84) and an R ­ 2 of 0.95 among women. without, HIV infection was modest among women (AUC
The risk-score categories and corresponding HIV prevalence of 0.66 for both the development and validation datasets),
among men and women are shown in Supplementary Table and somewhat higher among men (AUC of 0.76 and 0.71
SIV. for the development and validation datasets, respectively).
Although our study highlights variation in the performance
of gender-specific algorithms, majority of the HIV-risk fac-
Discussion tors included in the final models were similar for both sexes.
Use of a single overall algorithm may, therefore, be appro-
Our study demonstrates that a HIV predictive risk-score priate and likely more feasible to implement in the field.
algorithm, derived from a set of sociodemographic and The risk-score algorithm presented offers an evidence-
behavioral characteristics, can be used to identify sub-pop- base to guide identification of outpatient sub-populations
ulations who have higher risk of HIV infection to whom with higher risk for HIV, to whom HIV testing could be
HIV testing could be targeted. Other studies, which have prioritized. Our study found that targeted HIV testing using
evaluated similar risk-score algorithms for targeting HIV the three highest risk-score categories in the overall algo-
testing, have been conducted in specific settings (STI clinics, rithm, would dramatically reduce (by about 75%) the num-
a methadone clinic and a blood donor center) in the United ber of patients tested; however, this approach would miss the
States [53, 59]. Although the Denver risk-score algorithm diagnosis of approximately 50% of HIV infected individuals
(also evaluated in the United States) has been widely vali- accessing health facilities, making the use of the algorithm
dated, including in general outpatient care settings [25, 26, inferior to universal testing. Even for the gender-specific
60], it was developed using data from STI clinic attendees algorithm among men, which had superior discrimination
[27, 61]. To our knowledge, our study is the first to develop performance as compared to the overall algorithm, targeted
and validate an HIV testing algorithm using data from the HIV testing using the three highest risk-score categories
general outpatient care setting, and the first of its kind to be would reduce the number of patients tested by one half, and
conducted in the Sub-Saharan Africa setting. miss the diagnosis of approximately 14% of HIV infected
Our risk-score algorithm consists of simple variables, individuals. The algorithm’s use should, therefore, be con-
which in our study were collected within a routine health sidered in settings where resource or other logistical con-
care delivery setting, demonstrating the feasibility of imple- straints necessitate targeted testing, and should be coupled
mentation. The overall final algorithm comprised the fol- with other HIV testing strategies recommended by the World
lowing variables: age category 35–39/40–44 years; occupa- Health Organization [10, 62].
tion (manual/domestic or trade/sales/service); marital status The predictors included in our risk-score algorithm are
(polygamous marriage, separated/divorced or widowed); in consistent with those shown in other studies to be associ-
the prior 12 months having ≥ 2 sexual partners or reporting ated with higher risk of HIV infection. The pattern of HIV
treatment for an STI; and having never been tested for HIV prevalence by age and sex is consistent with national surveys
or having a negative HIV test result > 12 months prior. This in Kenya [63]. Furthermore, several studies have shown that
algorithm accounted for a high proportion of the variability polygamous marriage [29, 30], widowed status [31–33], or
of HIV prevalence in our development ­(R2 0.89) and valida- separated/divorced status [32–34]; having multiple sexual
tion ­(R2 0.88) study populations. The algorithm’s ability to partners [31, 34, 35]; having a new sexual partner [31, 35];
discriminate between individuals with, and without, HIV having an STI [31, 35]; and uncircumcised status among
infection in the general outpatient setting was modest (AUC men [43–45] are associated with higher risk of HIV infec-
of 0.69 for both the development and validation datasets) tion. Some studies have shown an association between HIV
and comparably lower than the Denver HIV risk-score algo- risk and higher socioeconomic status/employment/having
rithm (AUC range of 0.75–0.85) [25, 27, 60]. This likely income [31, 64–67], others have shown an association with
reflects more widespread distribution of HIV-risk factors low socioeconomic status [68, 69], while others have dem-
among persons accessing health facilities in the setting of onstrated a mixed association [70, 71] or no association
a generalized HIV epidemic, although ways to improve the [72–75]. Although we did not assess socioeconomic status
discrimination performance of the overall algorithm should directly, we found manual/domestic or trade/sales/service
be explored. occupations were associated with higher risk of HIV infec-
Among women, the proportion of variability in HIV prev- tion, which might be explained by an unidentified interplay
alence accounted for by the final model/algorithm was high between source of income and behavior, including increased
­(R2 of 0.87 in the development and 0.95 in the validation opportunity for social interaction and travel. Our findings
datasets), and varied among men ­(R2 of 0.69 and 0.85 in the are also consistent with program data from western Kenya

13
AIDS and Behavior

which found that patients who had never been tested for be externally validated in other regions and settings, and the
HIV, or had a negative HIV test result > 12 months prior impact of its use evaluated.
were more likely HIV-positive [76]. Most patients (95%) had
been tested for HIV within the previous 12 months, reflect-
ing intensified HIV testing efforts to increase ART coverage
Conclusions
in the study region [77–80]. Although studies have shown
that alcohol use [39, 40], intimate partner violence [41], and
In summary, our study demonstrates that a HIV predictive
having sex in exchange for money/favors [81] are associated
risk-score algorithm, derived from a set of sociodemo-
with higher risk of HIV infection, these were not signifi-
graphic and behavioral characteristics, can be used to iden-
cant in our study; possibly owing to these variables being
tify sub-populations who have higher risk of HIV infection
under-reported or being less prevalent in our study popula-
to whom HIV testing could be targeted. The overall algo-
tion of general outpatient attendees. Although other studies
rithm’s ability to discriminate between individuals with, and
have demonstrated an association of race/ethnicity with HIV
without, HIV infection in the general outpatient setting was
infection [52], this association has not been shown by stud-
modest. Additionally, using the three highest risk-score cat-
ies conducted in Kenya and was not evaluated in our study.
egories in the overall algorithm to target HIV testing would
Behavioral risk data were collected by trained counselors
dramatically reduce (by about 75%) the number of patients
at a private space, to facilitate patient privacy and reduce
tested, but miss the diagnosis of approximately 50% of HIV
social desirability bias. However, comparison of our study’s
infected individuals accessing health facilities, making the
patient characteristics with results from the most recent
use of the algorithm inferior to universal testing. Therefore,
(2014) Kenya Demographic and Health Survey suggests
in settings where universal testing is not feasible, the risk-
patients might have under-reported certain variables. The
score algorithm offers an evidence-base to guide identifi-
survey reported that 1.7% of women and 22% of men in the
cation of patient sub-populations with higher HIV risk, to
study region use alcohol [82], suggesting that the proportion
whom HIV testing could be targeted. Further evaluation is
of patients in our study who reported having sex under the
needed to explore ways to improve the discrimination perfor-
influence of alcohol (2%) is likely an underestimate. Simi-
mance of the algorithm, to externally validate the algorithm
larly, whereas the survey results showed that nationally 7.8%
in other regions and settings, and to assess the impact of its
of women and 2.3% of men experience sexual violence [82],
use.
our study found that 2% of patients reported being coerced
to have sex in the prior 12 months, also likely an underes- Acknowledgements  This study was made possible by support from the
timate. The proportion of patients who reported having sex U.S. President’s Emergency Plan for AIDS Relief (PEPFAR) through
in exchange for money/favors (3%) in our study, is however, the United States Centers for Disease Control and Prevention (CDC),
Division of Global HIV and TB (DGHT). The authors would like to
comparable to the national survey findings [82].
express their gratitude to the health care workers, HIV testing coun-
Our study had several limitations. First, our algorithm selors and data management staff at the health facilities described in
did not include all potential predictor variables, as education this study, without whom this study would not have been possible.
level, condom use and having an HIV-positive sexual partner The authors would also like to thank the Kenya Ministry of Health and
the County Departments of Health for Kisumu, Homa Bay and Siaya
were not included; however, we believe that the majority
Counties for the support they provided and continue to provide for
of behaviors that have been demonstrated to be associated HIV testing services.
with higher HIV infection in our study setting were included.
Secondly, our study did not meet the sample size rule of Disclaimer  The findings and conclusions in this report are those of
ten outcome events per variable recommended for clinical the authors and do not necessarily represent the official position of
the Centers for Disease Control and Prevention and other funding
predictive model evaluation [46, 83, 84]. Furthermore, by
institutions.
stratifying our analysis by gender, the sample size reduced
further. However, studies that have evaluated the effect of Open Access  This article is licensed under a Creative Commons Attri-
the sample size recommendation have shown conflicting bution 4.0 International License, which permits use, sharing, adapta-
results [85–87], and further evaluation of the rule has been tion, distribution and reproduction in any medium or format, as long
as you give appropriate credit to the original author(s) and the source,
recommended [87, 88]. To minimize overfitting occasioned
provide a link to the Creative Commons licence, and indicate if changes
by a small sample size, our study incorporated the use of were made. The images or other third party material in this article are
Akaike information criterion for variable selection in the included in the article’s Creative Commons licence, unless indicated
step-wise regression model [57, 89]. Finally, although the otherwise in a credit line to the material. If material is not included in
the article’s Creative Commons licence and your intended use is not
development of our algorithm derives strength from using
permitted by statutory regulation or exceeds the permitted use, you will
data from five health facilities located across three coun- need to obtain permission directly from the copyright holder. To view a
ties, data used for external validation was from a facility copy of this licence, visit http://creat​iveco​mmons​.org/licen​ses/by/4.0/.
located in the same region. The algorithm should therefore

13
 AIDS and Behavior

References 16. National AIDS and STI Control Programme (NASCOP). Prelimi-
nary KENPHIA 2018 Report. Nairobi: NASCOP; 2020. https​://
phia.icap.colum​bia.edu/wp-conte​nt/uploa​ds/2020/02/KENPH​
1. Joint United Nations Programme on HIV/AIDS. 90–90–90:
IA-2018_Preli​minar ​y-Repor ​t_final​-web.pdf. Accessed 9 May
An ambitious treatment target to help end the AIDS epidemic.
2020.
Geneva: Joint United Nations Programme on HIV/AIDS; 2014.
17. Bandasona T, McHugha G, Dauyaa E, et al. Validation of a screen-
https​://www.unaid​s.org/sites​/defau​lt/files​/media​_asset​/90-90-
ing tool to identify older children living with HIV in primary care
90_en.pdf. Accessed 5 Jun 2019.
facilities in high HIV prevalence settings. AIDS. 2016;30:779–85.
2. Joint United Nations Programme on HIV/AIDS. Fast-Track: end-
18. Ferrand R, Weiss H, Nathoo K, et al. A primary care level algo-
ing the AIDS epidemic by 2030. Geneva: Joint United Nations
rithm for identifying HIV-infected adolescents in populations at
Programme on HIV/AIDS; 2014. https​://www.unaid​s.org/sites​
high risk through mother-to-child transmission. Trop Med Int
/defau​l t/files​/ media​_ asset​/ JC268​6 _WAD20​1 4rep​o rt_en.pdf.
Health. 2011;30(3):349–55.
Accessed 5 Jun 2019.
19. Horwood C, Liebeschuetz S, Blaauw D, Cassol S, Qazi S.
3. Dieffenbach CW. Preventing HIV transmission through antiret-
Diagnosis of paediatric HIV infection in a primary health care
roviral treatment-mediated virologic suppression: aspects of an
setting with a clinical algorithm. Bull World Health Organ.
emerging scientific agenda. Curr Opin HIV AIDS. 2012;7:106–10.
2003;81(12):858–66.
4. Cohen MS, Chen YQ, McCauley M, et al. Prevention of HIV-1
20. Allison W, Kiromat M, Vince J, Handan C, Graham S, Kaldor
infection with early antiretroviral therapy. N Engl J Med.
J. Development of a clinical algorithm to prioritise HIV testing
2011;365:493–505.
of hospitalised paediatric patients in a low resource moderate
5. Cohen MS, Chen YQ, McCauley M, et al. Antiretroviral ther-
prevalence setting. Arch Dis Child. 2011;96(1):67–72.
apy for the prevention of HIV-1 transmission. N Engl J Med.
21. Bahwere P, Piwoz E, Joshua MC, et al. Uptake of HIV test-
2016;375(9):830–9.
ing and outcomes within a Community-based Therapeutic Care
6. The INSIGHT START Study Group. Initiation of antiretroviral
(CTC) programme to treat severe acute malnutrition in Malawi:
therapy in early asymptomatic HIV infection. N Engl J Med.
a descriptive study. BMC Infect Dis. 2008;8(1):106.
2015;373:795–807.
22. Temprano ANRS 12136 Study Group. A trial of early antiretro-
7. Joint United Nations Programme on HIV/AIDS. Global HIV and
virals and isoniazid preventive therapy in Africa. N Engl J Med.
AIDS statistics—2019 fact sheet. Geneva: Joint United Nations
2015;373(9):808–22.
Programme on HIV/AIDS; 2019. https:​ //www.unaids​ .org/en/resou​
23. Grinsztejn B, Hosseinipour MC, Ribaudo HJ, et al. Effects of
rces/fact-sheet​. Accessed 5 Jun 2019.
early versus delayed initiation of antiretroviral treatment on
8. Joint United Nations Programme on HIV/AIDS. UNAIDS data
clinical outcomes of HIV-1 infection: results from the phase
2019. Geneva: Joint United Nations Programme on HIV/AIDS;
3 HPTN 052 randomised controlled trial. Lancet Infect Dis.
2019. https​://www.unaid​s.org/sites​/defau​lt/files​/media​_asset​
2014;14(4):281–90.
/2019-UNAID​S-data_en.pdf. Accessed 5 Jun 2019.
24. World Health Organization. Consolidated guidelines on the use
9. President’s Emergency Plan for AIDS Relief. President’s Emer-
of antiretroviral drugs for treating and preventing HIV infection.
gency Plan for AIDS Relief 2019 Annual Report. Washington,
Geneva: World Health Organization; 2015. https​://apps.who.
DC: Office of the Global AIDS Coordinator; 2019.
int/iris/bitst​ream/handl​e/10665​/19806​4/97892​41509​893_eng.
10. World Health Organization. Consolidated guidelines on HIV
pdf?seque​nce=1. Accessed 11 May 2020.
testing services. Geneva: World Health Organization; 2015. https​
25. Haukoos J, Hopkins E, Bucossi M, et al. Validation of a quan-
://www.who.int/hiv/pub/guide ​ l ines ​ / hiv-testi ​ n g-servi ​ c es/en/.
titative HIV risk prediction tool using a national HIV testing
Accessed 12 Jun 2019.
cohort. J Acquir Immune Defic Syndr. 2015;68(5):599.
11. Kenya Ministry of Health. Kenya HIV testing guidelines. Nairobi:
26. Haukoos J, Hopkins E, Bender B, Sasson C, Al-Tayyib A,
Kenya Ministry of Health; 2015. https​://www.nasco​p.or.ke/wp-
Thrun M. Comparison of enhanced targeted rapid HIV screen-
conte​nt/uploa​ds/2016/08/THE-KENYA​-HIV-TESTI​NG-SERVI​
ing using the Denver HIV risk score to nontargeted rapid HIV
CES-GUIDE​LINES​.pdf. Accessed 12 Jun 2019.
screening in the emergency department. Ann Emerg Med.
12. Government of the Kingdom of Eswatini. Swaziland HIV Inci-
2013;61(3):353–61.
dence Measurement Survey 2 (SHIMS2) 2016–17: Final Report.
27. Haukoos J, Lyons M, Lindsell C, et al. Derivation and validation
Mbabane: Government of the Kingdom of Eswatini; 2018. https​
of the Denver Human Immunodeficiency Virus (HIV) risk score
://phia.icap.colum​bia.edu/wp-conte​nt/uploa​ds/2019/05/SHIMS​
for targeted HIV screening. Am J Epidemiol. 2012;175(8):838–46.
2_Final​-Repor​t_05.03.2019_forWE​B.pdf. Accessed 9 May 2020.
28. Powers KA, Miller WC, Pilcher CD, et al. Improved detection of
13. Ministry of Health and Social Services (MoHSS) Namibia.
acute HIV-1 infection in Sub-Saharan Africa: development of a
Namibia Population-based HIV Impact Assessment (NAMPHIA)
risk score algorithm. AIDS. 2007;21(16):2237.
2017: Final Report. Windhoek: MoHSS; 2019. https​://globa​lheal​
29. Adeokun LA, Nalwadda RM. Serial marriages and AIDS in
thsci​ences​.ucsf.edu/sites​/globa​lheal​thsci​ences​.ucsf.edu/files​/pub/
Masaka District. Health Transit Rev. 1997;7:49–66.
namph​ia-final​-repor​t_for-web.pdf. Accessed 9 May 2020.
30. Bove R, Valeggia C. Polygyny and women’s health in Sub-Saharan
14. Ministry of Health Lesotho, Centers for Disease Control and
Africa. Soc Sci Med. 2009;68(1):21–9.
Prevention (CDC), and ICAP at Columbia University. Lesotho
31. Amornkul PN, Vandenhoudt H, Nasokho P, et al. HIV prevalence
Population-based HIV Impact Assessment (LePHIA) 2016–2017:
and associated risk factors among individuals aged 13–34 years
Final Report. Maseru, Atlanta, New York: Ministry of Health,
in rural western Kenya. PLoS ONE. 2009;4(7):e6470.
CDC, and ICAP; 2019. https​://phia.icap.colum​bia.edu/wp-conte​
32. Tenkorang EY. Marriage, widowhood, divorce and HIV
nt/uploa​ds/2019/09/LePHI​A_Final​Repor ​t_Web.pdf. Accessed 9
risks among women in Sub-Saharan Africa. Int Health.
May 2020.
2014;6(1):46–53.
15. Ministry of Health Uganda. Uganda Population-based HIV Impact
33. Oluoch T, Mohammed I, Bunnell R, et al. Correlates of HIV infec-
Assessment (UPHIA) 2016–2017: Final Report. Kampala: Min-
tion among sexually active adults in Kenya: a national population-
istry of Health; 2019. https​://phia.icap.colum​bia.edu/wp-conte​
based survey. Open AIDS J. 2011;5:125.
nt/uploa​ds/2019/07/UPHIA​_Final​_Repor​t_Revis​e_07.11.2019_
34. Kimanga DO, Ogola S, Umuro M. Prevalence and incidence of
Final​_for-web.pdf. Accessed 9 May 2020.
HIV infection, trends, and risk factors among persons aged 15–64

13
AIDS and Behavior

years in Kenya: results from a nationally representative study. J 53. Gerbert B, Bronstone A, McPhee S, Pantilat S, Aller-
Acquir Immune Defic Syndr. 2014;66(Suppl 1):S13. ton M. Development and testing of an HIV-risk screening
35. Pettifor AE, Rees HV, Kleinschmidt I, et  al. Young people’s instrument for use in health care settings. Am J Prev Med.
sexual health in South Africa: HIV prevalence and sexual behav- 1998;15(2):103–13.
iors from a nationally representative household survey. AIDS. 54. Lazzarin A, Saracco A, Musicco M, Nicolosi A. Man-to-woman
2005;19(14):1525–34. sexual transmission of the human immunodeficiency virus: risk
36. Dunkle KL, Jewkes RK, Brown HC, Gray GE, McIntryre JA, Har- factors related to sexual behavior, man’s infectiousness, and wom-
low S. Transactional sex among women in Soweto, South Africa: an’s susceptibility. Arch Intern Med. 1991;151(12):2411–6.
prevalence, risk factors and association with HIV infection. Soc 55. Sauerbrei W, Royston P, Binder H. Selection of important
Sci Med. 2004;59(8):1581–92. variables and determination of functional form for continu-
37. Pettifor AE, Kleinschmidt I, Levin J, et al. A community-based ous predictors in multivariable model building. Stat Med.
study to examine the effect of a youth HIV prevention interven- 2007;26(30):5512–28.
tion on young people aged 15–24 in South Africa: results of the 56. Collins S, Ogundimu O, Cook J, Manach Y, Altman D. Quantify-
baseline survey. Trop Med Int Health. 2005;10(10):971–80. ing the impact of different approaches for handling continuous
38. Gelmon L. Kenya HIV prevention response and modes of trans- predictors on the performance of a prognostic model. Stat Med.
mission analysis. Nairobi: National AIDS Control Council; 2009. 2016;35(23):4124–35.
39. Zablotskaa II, Ronald HG, David S, et al. Alcohol use before 57. Cowley LE, Farewell DM, Maguire S, Kemp AM. Methodo-
sex and HIV acquisition: a longitudinal study in Rakai, Uganda. logical standards for the development and evaluation of clinical
AIDS. 2006;20:1191–6. prediction rules: a review of the literature. Diagn Progn Res.
40. Kalichman SC, Simbayi L, Jooste S, Vermaak R, Cain D. 2019;3(1):16.
Sensation seeking and alcohol use predict HIV transmis- 58. R Core Team. R: a language and environment for statistical com-
sion risks: prospective study of sexually transmitted infec- puting. Vienna: R Foundation for Statistical Computing; 2013.
tion clinic patients, Cape Town, South Africa. Addict Behav. https​://www.R-proje​ct.org/. Accessed 16 May 2020.
2008;33(12):1630–3. 59. Chen Z, Branson B, Ballenger A, Peterman T. Risk assessment to
41. Annie MD. Spousal intimate partner violence is associated with improve targeting of HIV counseling and testing services for STD
HIV and other STIs among married Rwandan women. AIDS clinic patients. Sex Transm Dis. 1998;25(10):539–43.
Behav. 2011;15:142–52. 60. Hsieh Y, Haukoos J, Rothman R. Validation of an abbreviated ver-
42. Eshleman SH, Hudelson SE, Redd AD, et  al. Treatment as sion of the Denver HIV risk score for prediction of HIV infection
prevention: characterization of partner infections in the HIV in an urban ED. Am J Emerg Med. 2014;32(7):775–9.
Prevention Trials Network 052 Trial. J Acquir Immune Defic 61. Rosenberg E, Delaney K, Branson B, Spaulding A, Sullivan P,
Syndr. 2017;74(1):112. Sanchez T. Re: “Derivation and validation of the Denver human
43. Auvert B, Taljaard D, Lagarde E, Sobngwi-Tambekou J, Sitta immunodeficiency virus (HIV) risk score for targeted HIV screen-
R, Puren A. Randomized, controlled intervention trial of male ing”. Am J Epidemiol. 2012;176(6):567–8.
circumcision for reduction of HIV infection risk: The ANRS 62. World Health Organization. Guidelines on HIV self-testing and
1265 Trial. PLoS Med. 2005;2(11):e298. partner notification: supplement to consolidated guidelines on
44. Bailey RC, Moses S, Parker CB, et al. Male circumcision for HIV testing services. Geneva: World Health Organization; 2016.
HIV prevention in young men in Kisumu, Kenya: a randomised https​://www9.who.int/hiv/pub/self-testi​ng/hiv-self-testi​ng-guide​
controlled trial. Lancet. 2007;369:643–56. lines​/en/. Accessed 29 Jan 2020.
45. Wawer MJ, Makumbi F, Kigozi G, et al. Circumcision in HIV- 63. National AIDS STI Control Programme Ministry of Health Kenya.
infected men and its effect on HIV transmission to female part- Kenya AIDS indicator survey 2012. Nairobi: Ministry of Health
ners in Rakai, Uganda: a randomised controlled trial. Lancet. Nairobi; 2013. https​://nacc.or.ke/wp-conte​nt/uploa​ds/2015/10/
2009;374:229–37. KAIS-2012.pdf. Accessed 29 Jan 2020.
46. Laupacis A, Sekar N. Clinical prediction rules: a review and 64. Msisha WM, Kapiga SH, Earls F, Subramanian SV. Socioeco-
suggested modifications of methodological standards. JAMA. nomic status and HIV seroprevalence in Tanzania: a counterintui-
1997;277(6):488–94. tive relationship. Int J Epidemiol. 2008;37(6):1297–303.
47. Wasson J, Sox H, Neff R, Goldman L. Clinical prediction rules: 65. Barongo LR, Borgdorff MW, Mosha FF, et al. The epidemiology
applications and methodological standards. N Engl J Med. of HIV-1 infection in urban areas, roadside settlements and rural
1985;313(13):793–9. villages in Mwanza Region, Tanzania. AIDS. 1992;6(12):1521–8.
48. Toll D, Janssen K, Vergouwe Y, Moons K. Validation, updating 66. Chao A, Bulterys M, Musanganire F, et al. Risk factors associ-
and impact of clinical prediction rules: a review. J Clin Epide- ated with prevalent HIV-1 infection among pregnant women in
miol. 2008;61(11):1085–94. Rwanda. Int J Epidemiol. 1994;23(2):371–80.
49. Moons K, Altman D, Reitsma J, et al. Transparent Reporting 67. Kapiga SH, Lyamuya EF, Vuylsteke B, Spiegelman D, Larsen U,
of a multivariable prediction model for Individual Prognosis or Hunter DJ. Risk factors for HIV-1 seroprevalence among family
Diagnosis (TRIPOD): explanation and elaboration. Ann Intern planning clients in Dar es Salaam, Tanzania. Afr J Reprod Health.
Med. 2015;162(1):W1–W73. 2000;4(1):88–99.
50. Kenya National AIDS Control Council. Kenya HIV Estimates 68. Farmer P. Infections and inequalities: the modern plagues. Berke-
Report. Nairobi: Kenya Ministry of Health; 2018. https​://nacc. ley: University of California Press; 2001.
or.ke/wp-conte​nt/uploa​ds/2018/11/HIV-estim​ates-repor​t-Kenya​ 69. Krueger LE, Wood RW, Diehr PH, Maxwell CL. Poverty and
-20182​.pdf. Accessed 23 Feb 2019. HIV seropositivity: the poor are more likely to be infected. AIDS.
51. Kenya National AIDS and STI Control Programme. HIV Pro- 1990;4(8):811–4.
gram Data. Nairobi: Kenya National AIDS and STI Control 70. Hargreaves JR, Morison LA, Chege J, et al. Socioeconomic status
Programme; 2017. and risk of HIV infection in an urban population in Kenya. Trop
52. Dube B, Marshall T, Ryan R, Omonijo M. Predictors of human Med Int Health. 2002;7(9):793–802.
immunodeficiency virus (HIV) infection in primary care among 71. Wojcicki JM. Socioeconomic status as a risk factor for HIV infec-
adults living in developed countries: a systematic review. Syst tion in women in East, Central and Southern Africa: a systematic
Rev. 2018;7(1):82. review. J Biosoc Sci. 2005;37(1):1–36.

13
 AIDS and Behavior

72. Moses S, Muia E, Bradley JE, et al. Sexual behaviour in Kenya: 2014. https:​ //dhspro​ gram.​ com/pubs/pdf/fr308/​ fr308.​ pdf. Accessed
implications for sexually transmitted disease transmission and 3 Feb 2020.
control. Soc Sci Med. 1994;39(12):1649–56. 83. Peduzzi P, Concato J, Kemper E, Holford T, Feinstein A. A simu-
73. Ayisi JG, Van E, Anna M, et al. Risk factors for HIV infection lation study of the number of events per variable in logistic regres-
among asymptomatic pregnant women attending an antenatal sion analysis. J Clin Epidemiol. 1996;49(12):1373–9.
clinic in western Kenya. Int J STD AIDS. 2000;11(6):393–401. 84. Harrell FE Jr, Frank E, Lee K, Mark D. Multivariable prognos-
74. Ryder RW, Ndilu M, Hassig SE, et al. Heterosexual transmission tic models: issues in developing models, evaluating assumptions
of HIV-1 among employees and their spouses at two large busi- and adequacy, and measuring and reducing errors. Stat Med.
nesses in Zaire. AIDS. 1990;4(8):725–32. 1996;15(4):361–87.
75. Serwadda D, Wawer MJ, Musgrave SD, Sewankambo NK, Kaplan 85. Vittinghoff E, McCulloch C. Relaxing the rule of ten events
JE, Gray RH. HIV risk factors in three geographic strata of rural per variable in logistic and Cox regression. Am J Epidemiol.
Rakai District, Uganda. AIDS. 1992;6(9):983–9. 2006;165(6):710–8.
76. Joseph R, Musingila P, Miruka F, et al. Expanded eligibility 86. Courvoisier D, Combescure C, Agoritsas T, Gayet-Ageron A, Per-
for HIV testing increases HIV diagnoses—a cross-sectional neger T. Performance of logistic regression modeling: beyond the
study in seven health facilities in western Kenya. PLoS ONE. number of events per variable, the role of data structure. J Clin
2019;14(12):e0225877. Epidemiol. 2011;64(9):993–1000.
77. US President’s Emergency Plan for AIDS Relief. Strategy for 87. Smeden M, Groot J, Moons K, et al. No rationale for 1 variable per
Accelerating HIV/AIDS Epidemic Control (2017–2020). Wash- 10 events criterion for binary logistic regression analysis. BMC
ington, DC: US President’s Emergency Plan for AIDS Relief; Med Res Methodol. 2016;16(1):163.
2017. https:​ //www.state.​ gov/wp-conten​ t/upload​ s/2019/08/PEPFA​ 88. Smeden M, Moons K, Groot J, et al. Sample size for binary logis-
R-Strat​egy-for-Accel​erati​ng-HIVAI​DS-Epide​mic-Contr​ol-2017- tic prediction models: beyond events per variable criteria. Stat
2020.pdf. Accessed 3 Feb 2020. Methods Med Res. 2019;28(8):2455–74.
78. United States President’s Emergency Plan for AIDS Relief. PEP- 89. Royston P, Moons K, Altman D, Vergouwe Y. Prognosis and
FAR Annual Report. 2017. prognostic research: developing a prognostic model. BMJ.
79. United States President’s Emergency Plan for AIDS Relief. PEP- 2009;338:b604.
FAR Annual Report. 2015.
80. United States President’s Emergency Plan for AIDS Relief. PEP- Publisher’s Note Springer Nature remains neutral with regard to
FAR Annual Report. 2016. jurisdictional claims in published maps and institutional affiliations.
81. Astemborski J, Vlahov D, Warren D, Solomon L, Nelson K.
The trading of sex for drugs or money and HIV seropositiv-
ity among female intravenous drug users. Am J Public Health.
1994;84(3):382–7.
82. Kenya National Bureau of Statistics. Kenya Demographic and
Health Survey 2014. Nairobi: Kenya National Bureau of Statistics;

13