Predicting Adult Stature for Individuals by A. F. Roche, H. Wainer, and D. Thissen. (review)

Predicting Adult Stature for Individuals. By A. F. Roche, H. Wainer, and D. Thissen. Basel: S. Karger, 1975. Pp. 114. $20.50. This attractive, soft-covered volume in the clean style that has become a Karger trademark is the third in a series entitled "Monographs in Paediatrics." Its goal is to present in great detail the new RWT (Roche-Wainer-Thissen) method of predicting adult height for children. The initial chapters discuss in detail the strengths and weaknesses of previous methods of height prediction before leading us through the genesis, statistics, and application of the RWT method. This work is the culmination of 15 years of study and research by an acknowledged authority on the topic, A. F. Roche of the FeIs Research Institute in Yellow Springs, Ohio. The population from which their new prediction formulas were derived consists of 800 individuals in the vicinity of the FeIs Institute who were followed closely from birth until 50 percent or more of their bones had attained adult configuration. Seventy-eight variables, including bone ages from the standard Greulich and PyIe atlas, radiographic measurements ofbone length and maturation , weight, recumbent length, and parent height, were all tested for predictive value. Of these, 18 measurements were found to correlate strongly with adult stature, and 14 of the 18 were related to radiographic appearance of various bones. These factors they further condensed into a "parsimonious set of predictors " suitable for clinical use that include (1) hand-wrist skeletal age, (2) average height of parents, (3) recumbent length, and (4) weight. The "smoothing" of relative predictor weightings is well presented and becomes necessary because the accuracy of any one predictive factor is not constant at all ages. At last, by page 68, we arrive at the "Final Predictive Tables" of the RWT method. There are separate predictive tables for both males and females that give "weighting" values for each of the four measurements made by the physician . By entering the table with the child's age to the nearest month, we find a row of four- or five-digit weighting factors by which the measured values must be multiplied. The products ofweight, recumbent length, bone age, and parents' average height, each multiplied by its own weighting factor, are then added with care to preserve the assigned plus or minus signs. A final constant obtained from the last column of the table is then added or substracted to give the clinician a final prediction of the child's adult height. It is quite straightforward and perhaps easier to perform than it is to describe. This is worked out in a single representative problem on page 83, but gadzooks, the numbers don't work! The difficulty seems to be that the constant (ß) used in their example is not the one they supply in the table! The error is small, but I found it very disconcerting since it is the only concrete example of their new method in use. Some of the symbols and equations are a bit confusing, too. On page 41, for example, an equation uses Y but the explanations discusses y ¦ Also, many of the symbols are not redefined at the bottom of each table so that the forgetful reader or one who has occasion to look at a single table will remain baffled unless he flips back to table 15, where abbreviations are explained. Even here the abbreviations could be more clearly represented; /3 appears in the explanation but not in the table because of a rather peculiar way of listing the definitions of related 442 I Book Reviews symbols. Also, one has to be very careful not to confuse the boys' table with that for girls. The tables run several pages each and are not reidentified on each new page. Just for fun, I used some numbers derived from one of my own children who has moderately retarded growth resulting from a hereditary blood dyscrasia and 9 years of intermittent prednisone therapy. The results were wildly optimistic, just as they are when his numbers are plugged into the older Bayley-Pinneau tables for predicting height. The error is less with the RWT method but still large. I always worry about midparent statures in...