Skip to main content

Table 3 Sensitivity, specificity, positive and negative predictive values of imputation method in the Hlabisa cohort

From: Imputing HIV treatment start dates from routine laboratory data in South Africa: a validation study

Specificity

Truly initiated ART = no

HGB/ALT + CD4

HGB/ALT only

Ever had an ART workup?

Yes

79

84

No

35,556

35,551

Total

35,635

35,635

Specificity

99.8%

99.8%

Positive Predictive Value (PPV)

Had ART workup = yes

HGB/ALT + CD4

HGB/ALT only

Known ART start date in 6mo after ART workup date?

Yes

17,970

18,393

No

866

963

Total

18,836

19,329

PPV

95.4%

95.2%

Negative Predictive Value (NPV)

Had ART workup = no

HGB/ALT + CD4

HGB/ALT only

Ever initiated ART?

Yes

3009

2,521

No

35,556

35,551

Total

38,565

38,072

NPV

92.2%

93.4%

  1. Table displays estimates in separate columns for our primary definition of ART workup: a patient’s first haemoglobin or ALT occurring up to 12 months after or one month before a CD4 count – and our secondary definition: first haemoglobin or ALT, regardless of CD4 count. For sensitivity and positive predictive value, an imputed ART start date and known start date were considered to “match” if the imputed date occurred in the interval 5 months and 1 week prior to and up to 3 weeks after the known ART start date. Because we imputed by adding 21 days to the date of the lab workup, a “match” occurred if the workup date was in the six months prior to known start date. For specificity and negative predictive value, there was no reference date – of initiation or lab workup – with which to define a six-month interval. Therefore, we report specificity and NPV for whether a patient ever had workup or initiated ART, respectively. We note that these are lower bounds for sensitivity and NPV, which would be higher were we to assign fake dates of initiation or workup to patients without them. Data are presented for Hlabisa only, as pre-ART data were not systematically collected in the RTC cohort