MISCELLANEA. 
I. A Note on the Age Distribution of Deaths from Diabetes Mellitus. 
By G. D. MAYNARD. 
In a paper entitled "A Statistical Study in Cancer Death-rates," published in Biometrika, 
Vol. vii. Part 3, brief mention was made of the fact that the Diabetes death-rates at age 
periods could not be fitted satisfactorily by any single frequency curve.. The form of the curve 
obtained seemed to indicate that the population dealt with under the term Diabetes Mellitus 
was not homogeneous. 
I have attempted therefore to resolve the complex distribution into its components. After 
some considerable labour, on the principle of trial and error, three curves have been found that 
give a very close fit to the original observations. For the sake of convenience I have termed the 
first sub-curve the Growth Curve, as its range is from birth to about 20 years ; this curve is 
symmetrical having its mean at 10 years, and being platykurtic belongs therefore to Type II. 
The second or Reproductive Curve approximately corresponds to the period of reproduction, 
that is from 15 to 50 ; this curve is also symmetrical and platykurtic, except in the case of the 
U.S.A. data where it was best fitted by the normal curve. The third or Old-age Curve is skew 
and of Type I. 
Finding, as might be expected, that the irregularities due to random sampling were of 
considerable inconvenience when dealing with only one year's returns, the figures dealt with in 
the paper above referred to were abandoned in favour of five-yearly returns for the period 
1900 — 1904. Three tables were available, England and Wales Male, England and Wales Female, 
and U.S.A. Male and Female ; separate tables for males and females in the case of U.S.A. were 
not obtainable for a five-yearly period. These data were then reduced to rates per million at 
each five-yearly age period. The higher rates observed in the U.S.A. table are largely due to 
the fact that the returns used are obtained chiefly from urban areas, whereas the English 
figures are compiled from the whole country. 
It will be seen that the various sub-curves agree very closely with each other, both as to 
range and position of the mean, but the final test as to the suitability of the resolution rests in 
the result of the application of the test for ' Goodness of Fit.' A total theoretical distribution 
was then constructed, by adding together the areas of the various components standing on the 
same five-yearly bases. The value of % 2 was then found and P, the probability of goodness of fit, 
obtained from the table in Biometrika, Vol. i. In both the English distributions the values of P 
could hardly be improved. In the U.S.A. figures the result is not quite so satisfactory when 
all the groups are taken into consideration. The number of cases of diabetes from which the 
rates were calculated is very small after the 80 — 85 age period and considerable variation may be 
expected ; in the 95 — 100 group only 3 cases were recorded in the 5 years. The effect therefore 
of a single case, or the misstatement of age in one case, will very materially alter the rate. 
I have therefore calculated x 2 f° r the first 19 and also 17 groups. The exclusion of the 20th 
group increases the value of P from - 6726 to - 9968, and if only the first 17 groups are taken 
P=-9994, which is I think a very satisfactory value. 
Medically this analysis is of some interest. Diabetes Mellitus has presented considerable 
difficulties and different types are well known. For instance, it has long been recognised that 
diabetes in the young adult is a much more serious disease than it is when it occurs in a person 
of advanced life. Then, again, pathologically there is the form associated with pancreatic disease, 
and a type in which hepatic disturbances are found. It is conceivable that the Growth Type 
may be associated with some pathological condition of an organ mainly of importance during 
development. The statistical analysis indicates that a different group of cause factors is 
Biometrika viii 29 
