A prediction risk score for HIV among adolescent girls and young women in South Africa: identifying those in need of HIV pre-exposure prophylaxis

Figures & data

Table 1. Characteristics of the sampled participants (N = 4,399).

Characteristic	Frequency	Weighted percentage
District
Cape Town	377	8.6
Ehlanzeni	803	18.3
O R Tambo	690	15.7
Tshwane	767	17.5
uThungulu	748	17.0
Zululand	1,014	22.1
Age group
15 – 19	2,515	57.2
20 – 24	1,884	42.8
Ever had sex
Yes	3,009	68.4
Age at sexual debut
≤ 15	259	5.9
16 – 24	2,750	62.5
Not applicable	1,390	31.6
Highest education
None or primary	210	4.8
Secondary	3,892	88.5
Some post-secondary	297	6.8
History of transactional sex
Yes	424	9.6
Socio-economic status
High SES	792	18.0
Low SES	3,607	82.0
Relationship status
Married	39	0.9
Dating	2,693	61.2
Single	1,624	36.9
Prefer not to say	43	1.0
HIV status
HIV positive	568	12.9
HIV negative	3,831	87.1
Any STI symptom
Yes	990	22.5
Partner HIV status
HIV positive	141	3.2
HIV negative	2,013	45.8
Prefer not to say	86	2.0
Do not know	899	20.4
Do not have partner	1,260	28.6
Often use substances
Yes	415	9.4
Ever pregnant
Yes	1,680	38.2
Ever raped
Yes	265	6.0

Table 2. Comparison of HIV status by predictor variables (N = 4,399).

Characteristic	HIV positive (Weighted row%)	HIV negative (Weighted row%)	P-Value
Age group (years)			<0.001
15 – 19	185 (6.75)	2,330 (93.25)
20 – 24	383 (19.7)	1,501 (80.3)
Age of sexual debut			<0.001
≤ 15	43 (15.9)	216 (84.1)
16 – 24	456 (15.25)	2294 (84.75)
Not Applicable	69 (4.5)	1.321(95.5)
Transactional sex			<0.001
Yes	90 (20.0)	334 (80.0)
No	478 (11.56)	3,497(88.44)
Socio-economic status			0.036
High SES	78 (9.20)	714 (90.80)
Low SES	490 (13.20)	4,117 (86.8)
Highest education			0.146
None or primary	35 (15.35)	175 (84.65)
Secondary	501 (12.36)	3,391 (87.64)
Some post-secondary	32 (10.31)	265 (89.69)
Partner older than 5yrs			<0.001
Yes	211 (20.28)	839 (78.72)
No	304 (13.92)	1,785 (86.08)
Do not know/NA	53 (3.69)	1,207 (96.31)
Partner HIV status			<0.001
HIV positive	84 (57.63)	57 (42.36)
HIV negative	244 (7.45)	3,029 (92.55)
Prefer not to say	21 (23.68)	65 (76.32)
Do not know	219 (24.24)	680 (75.76)
Ever treated for STIs			<0.001
Yes	191 (18.45)	799 (81.55)
No	377 (10.56)	3,032 (89.44)
Has more than six partners			0.7045
Yes	45 (11.67)	299 (88.33)
No	523 (12.43)	3,532 (87.57)
One parent dead			<0.001
Yes	345 (16.76)	1,688 (83.24)
No	219 (9.0)	2,065 (91.0)
Often use substances			0.5590
Yes	57 (12.14)	358 (87.86)
No	511 (12.39)	3,473 (87.61)
Ever pregnant			<0.001
Yes	338 (20.12)	1,342 (79.88)
No	223 (8.35)	2,447 (91.65)
Ever raped			0.003
Yes	50 (18.87)	215 (81.13)
No	518 (12.53)	3,616 (87.47)

Table 3. Coefficients and risk scoring for the independent predictors of HIV using the training dataset.

				95% CI
Predictor	Coefficient	Risk score	P-Value	Lower	Upper
Age group
15 – 19	0	0
20 – 24	0.7013	0.70	<0.001	0.3988	1.0039
Age at first sex
≤ 15	0
>15	−0.0371	−0.04	0.883	−0.5305	0.4562
Relationship status
Single	0.3084	0.31	0.051	−0.0015	0.6183
Dating	0	0
Married	−0.2600	−0.26	0.716	−1.6592	1.1392
Prefer not to say	0.7405	0.74	0.250	−0.5212	2.0023
Education
Primary	0	0
Secondary	−0.7093	−0.71	0.020	−0.3047	−0.1140
Post-secondary	−1.2398	−1.24	<0.001	−2.0175	−0.4620
Socioeconomic status
Relatively low SES	0.1701	0.17	0.369	−0.2007	0.5409
High SES	0	0
Partner 5 years older
Yes	0.1956	0.20	0.145	−0.0675	0.4587
No	0	0
No partner	−0.3038	−0.30	0.383	−0.9864	0.3837
Any STI symptom
Yes	0.3240	0.32	0.025	0.0416	0.6063
No	0	0
Partner HIV positive
No	0	0
Yes	2.5827	2.58	<0.001	2.1077	2.0583
Prefer not to say	1.3863	1.39	<0.001	0.6121	2.1605
Do not know	0.9275	0.93	<0.001	0.6568	1.1983
No partner	−1.558	1.56	0.056	−2.2035	0.0859
Used condom at last sex
Yes	0.1575	0.16	0.277	−0.1267	0.4419
No	0	0
Prefer not to say	−0.0197	−0.02	0.920	−0.4036	0.3641
Ever engaged in transactional sex
No	0	0
Yes	0.1818	0.18	0.328	−0.2064	0.5701
Parent dead
No	0	0
Yes	0.3190	0.32	0.016	0.0603	0.5778
Ever pregnant
No	0	0
Yes	0.1209	0.12	0.418	−0.1720	0.4139
Ever raped
No	0	0
Yes	0.1285	0.13	0.621	−0.3813	0.6384

Table 4. Performance characteristics of the selected risk scores based on the training dataset.

Cut point	Sensitivity	Specificity	Correctly classified	LR+	LR-
> =0	100.00%	0.00%	12.91%	1.0000
> =0.5	100.00%	0.03%	12.93%	1.0003	0.0000
> =1.0	100.00%	0.57%	13.41%	1.0058	0.0000
> =1.5	98.59%	7.60%	18.35%	1.0670	0.1854
> =2.0	88.03%	33.12%	40.21%	1.3163	0.3614
> =2.5	61.09%	69.15%	68.11%	1.9800	0.5627
> =3.0	38.56%	84.23%	78.34%	2.4455	0.7294
> =3.5	12.32%	96.95%	86.02%	4.0353	0.9044
> =4.0	0.00%	100.00%	87.098		1.0000

The optimal cut-off point of the risk score was 2.43 with a sensitivity of 71% and specificity of 60%.

Key.

Sensitivity: The proportion of actual positives which are correctly identified as such.

Specificity: The proportion of negatives which are correctly identified as such.

LR+: The ratio of the probability of a positive test among the truly positive subjects to the probability of a positive test among the truly negative subjects.

LR-: The ratio of the probability of a negative test among the truly positive subjects.

Figure 1. Training and testing AUC of the prediction model.

X-axis is the model’s sensitivity while y-axis is 1–Specificity. The blue line represents the testing ROC while red line represents the training ROC.

Table 5. Prediction model’s classification table based on the training dataset.

Sensitivity	16.67%
Specificity	98.48%
Positive predictive value	68.24%
Negative predictive value	85.77%
False positive rate	1.52%
False negative rate	83.33%
False positive rate for classified positive	31.76
False negative rate for classified negative	14.23
Correctly classified	85.07

Table 6. Performance of the risk prediction model on the training and testing datasets.

	Training set		Testing set
	Value	Remark	Value	Remark
Discrimination
AUC	0.778	Good	0.764	Good
Calibration
Hosmer Lemeshow	0.5638	Good	0.3627	Good
Brier Score	0.0954	Good	0.0954	Good
Pseudo R^2	0.1502	Good	0.1727	Good

Availability of data and materials

The datasets analyzed during the current study are not publicly available due

Data protection policy of South African Medical Research Council (SAMRC) which only makes the data available on request. Data can be requested from SAMRC through Dr Darshini Govindasamy on [email protected]