Role of point of care Hb diagnostic devices in getting the right picture of anemia control: Tangi Rural Anemia Diagnostic Accuracy Study

Figures & data

Figure 1. STARD diagram reporting flow of participants through the study comparing HC201, HCS with SYSMEX XP100 for Hb estimation in rural population of Odisha, India 2017.

Figure 2. Age and sex distribution of subjects undergoing diagnostic accuracy testing in Tangi Rural Anemia diagnostic accuracy study, 2017.

Figure 3. Box plots of population subgroups depicting distribution of true Hb values as determined by reference standard.

Table 1. Accuracy parameters of index test (HC₂₀₁) among study participants in Tangi Rural Anemia diagnostic accuracy study, 2017.

Statistic	All samples	Adolescent boys	Adolescent girls	Adult males	Adult females
HC201	(N = 1713)	(N = 406)	(N = 430)	(N = 398)	(N = 479)
	Value (95% CI)	Value (95% CI)	Value (95% CI)	Value (95% CI)	Value (95% CI)
Sensitivity	88.97% (85.52–91.84)	80.39% (66.88–90.18)	89.11% (81.35–94.44)	83.61% (71.91–91.85)	92.82% (88.25–96.02)
Specificity	78.93% (76.61–81.11)	77.46% (72.76–81.71)	75.68% (70.68–80.22)	86.94% (82.87–90.35)	75.00% (69.54–79.93)
Positive likelihood ratio	4.22 (3.78–4.71)	3.57 (2.82–4.52)	3.66 (2.99–4.49)	6.4 (4.76–8.62)	3.71 (3.02–4.56)
Negative likelihood ratio	0.14 (0.11–0.18)	0.25 (0.14–0.44)	0.14 (0.08–0.25)	0.19 (0.11–0.33)	0.1 (0.06–0.16)
Disease PREVALENCE	23.82% (21.82–25.91)	12.56% (9.50–16.18)	23.49% (19.56–27.79)	15.33% (11.93–19.25)	40.71% (36.27–45.26)
Positive predictive value	56.90% (54.17–59.58)	33.88% (28.82–39.35)	52.94% (47.88–57.94)	53.68% (46.27–60.94)	71.83% (67.49–75.79)
Negative predictive value	95.81% (94.55–96.79)	96.49% (94.02–97.96)	95.77% (92.81–97.54)	96.7% (94.32–98.10)	93.83% (90.14–96.20)
Accuracy	81.32% (79.39–83.14)	77.83% (73.47–81.78)	78.84% (74.67–82.60)	86.43% (82.67–89.64)	82.25% (78.53–85.57)

Table 2. Accuracy parameters for HCS at different cut-offs among study participants in Tangi Rural Anemia diagnostic accuracy study, 2017.

	HCS at clinical cut-offs	HCS <9	HCS <7	HCS <5
	Value (95% CI)	Value (95% CI)	Value (95% CI)	Value (95% CI)
HCS (N = 1728)
Sensitivity	99.27% (97.87–99.85)	71.88% (67.26–76.19)	22.98% (18.99–27.37)	2.44% (1.18–4.45)
Specificity	2.2% (1.48–3.14)	49.89% (47.15–52.62)	88.86% (87.03–90.50)	99.39% (98.81–99.74)
Positive likelihood ratio	1.01 (1.00–1.03)	1.43 (1.32–1.56)	2.06 (1.63–2.61)	4.03 (1.60–10.15)
Negative likelihood ratio	0.33 (0.10–1.09)	0.56 (0.48–0.66)	0.87 (0.82–0.92)	0.98 (0.97–1.00)
Disease prevalence	23.67% (21.68–25.75)	23.67% (21.68–25.75)	23.67% (21.68–25.75)	23.67% (21.68–25.75)
Positive predictive value	23.94% (23.73–24.15)	30.79% (29.09–32.54)	39.00% (33.60–44.69)	55.56% (33.18–75.88)
Negative predictive value	90.63% (74.75–96.93)	85.12% (82.92–87.08)	78.82% (77.86–79.74)	76.67% (76.38–76.95)
Accuracy	25.17% (23.14–27.29)	55.09% (52.71–57.46)	73.26% (71.11–75.34)	76.45% (74.37–78.43)
Adolescent boys HCS (N = 406)
Sensitivity	100% (93.02–100.00)	70.59% (56.17–82.51)	13.73% (5.70–26.26)	0.00% (0.00–6.98)
Specificity	0.56% (0.07–2.02)	42.82% (37.61–48.15)	86.20% (82.17–89.61)	98.87% (97.14–99.69)
Positive Likelihood Ratio	1.01 (1.00–1.01)	1.23 (1.01–1.51)	0.99 (0.48–2.07)	0.00
Negative Likelihood Ratio	0	0.69 (0.44–1.07)	1.00 (0.89–1.13)	1.01 (1.00–1.02)
Disease prevalence	12.56% (9.50–16.18)	12.56% (9.50–16.18)	12.56% (9.50–16.18)	12.56% (9.50–16.18)
Positive Predictive Value	12.62% (12.54–12.71)	15.06% (12.69–17.79)	12.50% (6.41–22.96)	0
Negative Predictive Value	100%	91.02% (86.69–94.03)	87.43% (86.08–88.66)	87.31% (87.19–87.44)
Accuracy	13.05% (9.93–16.73)	46.31% (41.37–51.29)	77.09% (72.69–81.09)	86.45% (82.73–89.63)
Adolescent girls HCS (N = 430)
Sensitivity	99.01% (94.61–99.97)	81.19% (72.19–88.28)	34.65% (25.46–44.77)	3.96% (1.09–9.83)
Specificity	0.91% (0.19–2.64)	21.58% (17.26–26.42)	77.81% (72.93–82.19)	98.78% (96.92–99.67)
Positive Likelihood Ratio	1 (0.98–1.02)	1.04 (0.93–1.16)	1.56 (1.12–2.18)	3.26 (0.83–12.79)
Negative Likelihood Ratio	1.09 (0.11–10.32)	0.87 (0.55–1.37)	0.84 (0.72–0.98)	0.97 (0.93–1.01)
Disease prevalence	23.49% (19.56–27.79)	23.49% (19.56–27.79)	23.49% (19.56–27.79)	23.49% (19.56–27.79)
Positive Predictive Value	23.47% (23.08–23.87)	24.12% (22.17–26.18)	32.41% (25.53–40.14)	50.00% (20.30–79.70)
Negative Predictive Value	75.00% (23.98–96.61)	78.89% (70.34–85.48)	79.50% (76.89–81.89)	77.01% (76.27–77.74)
Accuracy	23.95% (19.99–28.28)	35.58% (31.05–40.31)	67.67% (63.03–72.08)	76.51% (72.21–80.44)
Adult males HCS (N = 412)
Sensitivity	98.39% (91.34–99.96)	33.87% (22.33–47.01)	12.90% (5.74–23.85)	1.61% (0.04–8.66)
Specificity	3.71% (1.99–6.27)	89.43% (85.72–92.45)	100.00% (98.95–100.00)	100.00% (98.95–100.00)
Positive Likelihood Ratio	1.02 (0.98–1.06)	3.20 (2.02–5.09)	0	–
Negative Likelihood Ratio	0.43 (0.06–3.26)	0.74 (0.62–0.89)	0.87 (0.79–0.96)	0.98 (0.95–1.02)
Disease prevalence	15.05% (11.74–18.87)	15.05% (11.74–18.87)	15.05% (11.74–18.87)	15.05% (11.74–18.87)
Positive Predictive Value	15.33% (14.84–15.83)	36.21% (26.33–47.40)	100.00%	100.00%
Negative Predictive Value	92.86% (63.39–98.99)	88.42% (86.42–90.15)	86.63% (85.48–87.70)	85.16% (84.75–85.56)
Accuracy	17.96% (14.37–22.02)	81.07% (76.95–84.74)	86.89% (83.25–90.00)	85.19% (81.39–88.48)
Adult females HCS (N = 480)
Sensitivity	99.49% (97.18–99.99)	79.49% (73.13–84.92)	22.56% (16.90–29.08)	2.56% (0.84–5.88)
Specificity	3.86% (1.94–6.80)	42.81% (36.99–48.78)	91.23% (87.32–94.24)	100.00% (98.71–100.00)
Positive Likelihood Ratio	1.03 (1.01–1.06)	1.39 (1.23–1.57)	2.57 (1.63–4.06)	–
Negative Likelihood Ratio	0.13 (0.02–1.02)	0.48 (0.35–0.65)	0.85 (0.78–0.92)	0.97 (0.95–1.00)
Disease prevalence	40.62% (36.20–45.17)	40.62% (36.20–45.17)	40.62% (36.20–45.17)	40.62% (36.20–45.17)
Positive Predictive Value	41.45% (40.84–42.07)	48.74% (45.67–51.82)	63.77% (52.73–73.52)	100.00%
Negative Predictive Value	91.67% (58.88–98.83)	75.31% (69.17–80.57)	63.26% (61.29–65.19)	60.00% (59.45–60.55)
Accuracy	42.71% (38.24–47.27)	57.71% (53.15–62.17)	63.33% (58.85–67.65)	60.42% (55.88–64.82)

Figure 4. Bland Altman plot comparing HC201 with SYSMEX XP100 for Hb estimation in rural population of Odisha, India 2017.

Figure 5. Bland Altman plot comparing HCS with SYSMEX XP100 for Hb estimation in rural population of Odisha, India 2017.

Figure 6. ROC curve comparing HCS with HC201 for Hb estimation in rural population of Odisha, India 2017.

Table 3. ROC analysis parameters during comparison of HCS and HC₂₀₁ for Hb estimation in rural population of Odisha, India 2017.

POCDs	Proposed cut-off with maximum accuracy	Accuracy at proposed cut-off	AUC	Proportion of anemia cases in subsample undergone paired testing
Hemocue 201 (n = 1713)	11.4	0.852	0.887 (0.865–0.908)	23.82%
Hemoglobin Colour Scale (HCS) (n = 1713)	6	0.762	0.64 (0.612–0.669)	23.82%

Figure 7. Scatter of difference of Hb values estimated by HC201 and Reference test against true Hb values provided by the reference test.