In this paper, we introduce theory and practice method about combining independent studies of Diagnostic test into a summary ROC curve. This is a useful and easily applied by clinicians to analyse the data of diagnostic test. It’s referance for user and doer of EBM in China.
Objective To explore the accuracy and practicability of bone age assessment for the diagnosis of idiopathic precocious puberty (IPP). Methods According to the “Gold Standard”, we selected 55 girls with IPP for the study group, and 83 normal girls for the control group. We retrospectively analyzed the first left hand-wrist radiographs at the first visit. Bone ages were assessed by using a single-blind method according to the RUS (Radius Ulna and Short bones), carpale and 20 bones method (TW2). Each had 5 decision thresholds (gt;97th percentile, gt;90th percentile, gt;75th percentile, gt;50th percentile and ≤50th percentile). The diagnostic values from RUS, carpale and 20 bones methods assessing bone age were analyzed to identify the best decision threshold. Results ① Both sensitivity and specificity of the four decision thresholds were relatively higher, including gt;90th percentile of RUS (sensitivity 0.836, specificity 0.916), gt;90th percentile of carpale (sensitivity 0.746, specificity 0.916), gt;90th and gt;75th percentile of 20 bone (sensitivity 0.746, specificity 0.964 and sensitivity 0.982, specificity 0.783, respectively). ② Area under receiver operator characteristic curve (AUR): AUR of RUS 0.939 ± 0.019 (95%CI 0.902 to 0.977), AUR of carpale 0.899 ± 0.028 (95%CI 0.845 to 0.954), AUR of 20 bone 0.958 ± 0.014 (95%CI 0.930 to 0.986). No significant difference was found (F=2.03, P=0.13). ③ Agreement assessment within-observer reliability was 89.28%, and between-observer reliability was 80.3% (Kappa 0.68, u=6.87, P＜0.01). Conclusions RUS and 20 bones methods have high accuracy for the diagnosis of idiopathic precocious puberty. Considering sensitivity and specificity, we think that ＞90th percentile of RUS is the best decision threshold.
Objective To evaluate the diagnostic value of creatinine, as an indicator of glomerular filtration function injury. Methods MEDLINE, EMBASE and CBM-disc were searched from 1993 to 2003. Thirty four articles that described biomedical markers to indicate glomerular filtration function injury were selected according to specified inclusion criteria. These articles were evaluated systematically using SPSS, EXCEL, and RevMan software. Results The odds ratios of creatinine was 24.23. Areas under summarized receiver operator characteristic curve were 0.871. Selected articles were divided to groups for analysis according to diagnostic standards, such as inulin, iohexol, 125I-iothalamate, 51Cr-EDTA, 99mTc-DTPA and sodium thiosulfate. Inter-group analysis of creatinine was not statistically different (P=0.32). Intra-group analysis of inulin, 51Cr-EDTA, 99mTc-DTPA, and iohexol was not statistically different, P value were 0.61, 0.50, 0.36, 0.32, respectively. Intra-group analysis of 125I-iothalamate was statistically different (P=0.02). Selected articles were sub-grouped according to different analytic techniques of creatinine. Intra-group analysis by the Jaffe method was statistically different (P=0.03), intra-group analysis of enzymatic method was not statistically different (P=0.22). Conclusion The diagnostic value of creatinine was not qualified enough to indicate glomerular filtration function injury. Enzymatic methods are recommended to measure creatinine. Inulin or 51Cr-EDTA is suggusted to measure glomerular filtration rate in investigation of creatinine，and feedforward cohort is recommended to apply.
Objective To explore the solutions of problems with the ROC analysis for different types data. Method Two kinds of ROC analyses of three cardiac infarction markers, cTNT, CK-MB mass and MYO, were performed with the ROC program developed by Yunnan Provincial Clinical Laboratory Center. Results The distribution of prime data had a large range, which produced a bad analysis result. After logarithmic transformation, the prime data that had smaller range now can be analyzed with full-span method. The results were similar to the ROC analyzed with the overlapped data. Conclusions We should choose different statistical method depend on the distribution of data when we performed ROC analyses.
This paper introduced the fundamental theory, method advantages, application scenario and R software implementation method of the covariate-adjusted receiver operating characteristic (ROC) curve. Compared with the traditional univariate ROC curve, the covariate-adjusted ROC curve has distinct methodological advantages and wider application scenarios, which can help to evaluate the ability of markers to predict the targeted outcome more scientifically. It merits more widespread and prior adoption in practical research.