Stata se ver 14

We used adjusted odds ratios, relative risks, or hazard ratios (HRs) with 95% confidence intervals (CIs) from individual studies in order to calculate a pooled effect size. To measure heterogeneity across studies, we used Higgins I² [24 (link)], which is calculated as the following formula:
I²=100%×(Q–df)/Q
, where Q is Cochran’s heterogeneity statistic and df is the degrees of freedom. Negative values of I² are set at zero; I² ranges from 0% (no heterogeneity) to 100% (maximal heterogeneity). If I² value is greater than 50%, it represents the substantial heterogeneity [25 ]. Because individual studies were conducted in different populations, the random-effects model with the DerSimonian and Laird method was used to calculate the pooled effect size [26 (link),27 (link)].
Publication bias was assessed by using the Begg’s funnel plot and Egger’s test [28 (link)]. If the funnel plot is asymmetric or the p-value for Egger’s test is lower than 0.05, there exists publication bias. The Stata SE ver. 14.0 software (StataCorp., College Station, TX) was used for all statistical analyses.

Chi-square tests and analysis of variance with Tukey’s method for individual difference were used to assess the association between subject characteristics and USC types (i.e., no USC, usual place only, and usual doctor). Chi-square tests and Student t-tests were used to assess the association between subject characteristics and overall health care service quality. Multiple logistic regression analysis was used to assess the association between USC types and overall health care service quality after adjusting for socio-demographic and health status variables. First, those without a USC were set as a reference group (model I). Second, those having a place only were set as a reference group to assess for significant differences between those having a place only and those having a usual doctor (model IA). The Hosmer–Lemeshow test was applied to determine the goodness of fit of the logistic regression model. The discriminative ability of the model was assessed using the concordance (C) statistic, the area under the receiver operating characteristic curve. C statistics generally range from 0.5 (random concordance) to 1 (perfect concordance) [25 (link)]. Stata SE ver. 14.0 (Stata Corp., College Station, TX, USA) was used for statistical analysis, with P≤0.05 regarded as a significant difference. Sampling weights were applied in all analyses when calculating the P-values.