Forecasting of recall rate in the screening program

The recall of neonates with primary abnormal tests has almost always been considered the first step in the detection of congenital hypothyroidism (CH) in the CH screening programs. Considering recall rate as one of the indicators of cost-benefit and operational feasibility assessment of a program, forecasting of recall rate(s) could be beneficial to an optimized allocation of funds and resources. The trend of time-series of neonatal monthly recall rates was assessed in this report. From March 1998 to March 2002 (48 months), cord blood spot samples from 8 hospitals and a rural birth center in Tehran and Damavand District were collected on filter papers (Whatman BFC 180). TSH values were measured in the Endocrine Research Center laboratory using a two-site IRMA method. Primary TSH levels  20 mU/L were considered abnormal and recalled for serum confirmatory tests. Monthly recall rate (total number of primary TSH levels  20 mU/L to total live births per month  100) was recorded for 48 consecutive months. Monthly recall rates were analyzed using linear models of Box and Jenkins (ARIMA). The autoregressive model was fitted to forecast the next recall rates with a 95% confidence interval. Out of 29908 cord blood spot samples, 314 showed TSH levels  20 mU/L (total recall rate of 1.05%) of whom 32 had CH (1: 935 live births). Total data showed normal distribution and were stationary. Fitting of particular type of AR(3) model showed that monthly recall rate is predictable according to the recall rate obtained three months ago. Therefore, the autoregressive model of zt = 0.67 + 0.45 zt-3 + et, with significant parameters (p  0.003), was obtained to forecast the next recall rates, ultimately. Forecasting of monthly recall rates is possible using the autoregressive model.