Measurement of diet in a large national survey: comparison of computerized and manual coding of records in household measures

A diary method using household measures was employed to obtain dietary records in a large national prospective survey and a computer program, DIDO (Diet In Data Out), was designed for direct entry of the diaries. The accuracy of this computerized coding system was examined alongside that of the manual coding used for a similar diary in a previous wave, 7 years earlier, of the same survey. Accuracy was assessed by analysis of the errors in the coded and checked records by stringent re-checking of nominal 2% random subsamples of the diet diaries coded by each method. The mean time to code and check each of the 2086 7-day records in the whole survey using DIDO was 58 minutes (SD 30) compared with reported results of 1–4 hours for manual methods. The mean error rate of computerized coding and checking with DIDO was 2.3% (SD 2.1; range 0–8.9) per diary in the subsample. Correcting these mistakes made insignificant changes to the calculated mean energy and nutrient intakes for the subsample. The percentage of individuals changing to an adjacent third of nutrient distribution after correcting unambiguous errors ranged from none (for alcohol) to 11% (for carbohydrate and calcium intake). The mean error rate on a similar subsample of diaries from the earlier survey which had been coded manually was significantly higher at 5.9% (SD 4.1; range 0–17) per diary. Emphasis is laid on the importance, in coding, of dealing with ambiguities in the subjects' records, since this can affect the accuracy and the precision of the nutrient results obtained. We conclude that the DIDO coding method has the advantages of greater accuracy, speed, consistency and efficient data handling, and affords greater data accessibility for checking, compared with manual systems.