TABLE 1.
Diagnostic accuracy of Visitect CD4 at a cutoff of ≤350 cells/mm3 from GLMMa
| Factor | Value (95% CI) |
Inferenceb and heterogeneity |
|||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Model A |
Model B |
Model C |
|||||||||||||
| Sens | Spec | PPV | NPV | Sens | Spec | PPV | NPV | Sens | Spec | PPV | NPV | LR (χ2 ) | P | ICC | |
| Overall | 75.9 (66.1–85.8) | 84.3 (79.3–89.3) | 64.7 (53.1–76.3) | 90.5 (85.7–95.3) | 0.57 | ||||||||||
| Operator | χ2(4) = 7.48 | 0.113 | 0.58 | ||||||||||||
| Lab tech | 80.6 (67.9–93.3) | 84.8 (77.8–91.8) | 66.7 (52.8–80.5) | 92.4 (87.0–97.8) | |||||||||||
| Nurse | 70.5 (58.3–82.7) | 86.7 (81.2–92.2) | 67.7 (54.3–80.8) | 88.4 (82.5–94.3) | |||||||||||
| Counselor | 82.4 (70.1–94.8) | 79.7 (72.0–87.4) | 58.8 (45.3–72.4) | 93.1 (87.8–98.4) | |||||||||||
| Sample type | χ2(2) = 14.32 | 0.001 | 0.61 | ||||||||||||
| Venous | 81.7 (72.3–91.1) | 82.6 (77.1–88.1) | 63.0 (51.3–74.7) | 92.6 (88.0–97.1) | |||||||||||
| Finger-prick | 60.7 (45.0–76.3) | 89.5 (83.2–95.8) | 71.9 (56.2–87.5) | 85.0 (78.0–92.0) | |||||||||||
GLMM generalized through use of a logit link function and binomial distribution with a random intercept for test participant (n = 147). Sens, test sensitivity; Spec, test specificity; PPV, positive predictive value; NPV, negative predictive value; ICC, intraclass correlation coefficient from random intercept GLMM; model A, unadjusted; model B, operator by test interaction and main effect (not shown); model C, Blood sample type by test interaction and main effect (not shown). PPV and NPV estimates were determined from ordinary logit GLM with cluster robust standard errors. GLMM analyses would not converge reliably.
Likelihood ratio (LR) tests comparing nested less-constrained models (B and C) with model A.