Ratio of controls to cases: A design decision

SK Raina 
 Department of Community Medicine, Dr. Rajendra Prasad Government Medical College, Kangra, Himachal Pradesh, India

Correspondence Address:
Dr. S K Raina
Department of Community Medicine, Dr. Rajendra Prasad Government Medical College, Kangra, Himachal Pradesh
India

Full Text

Sir,

I have a few concerns on the article entitled "A prospective study of prevalence and association of peripheral neuropathy in newly diagnosed type 2 diabetes mellitus". [1] The authors state that over a period of 1.5 years (April 2010 to October 2011), 195 consecutive patients with newly diagnosed T2DM (mean age 47.6 ± 10.2 years, 59.0% males) were included in the study.Further, 75 healthy subjects (age 45.9 ± 9.9 years, 72% males) with normal fasting glucose levels, matched for age and sex with the patients, served as controls. The study aimed to establish an association between peripheral neuropathy and diabetes mellitus. The study design used by the authors (as appears from material and methods) is a case-control design with 195 cases and 75 controls. However, from a look at the results and analysis as conducted by the authors, it appears that the controls have not been accounted at all in the results. Agreed that the case-control design would have given us a better idea, but accounting for them is the most important step to establish association. What the authors actually have done is that they have reported on the prevalence of peripheral neuropathy among newly diagnosed diabetics (according to their definition) and then established an association between various characteristics of diabetics and peripheral neuropathy.

Further, fewer controls than cases have been chosen. The number of controls is an important design decision and authors need to have commented on it. It is useful to consider the ratio of controls to cases. There is usually little marginal increase in precision from increasing the ration of controls to cases beyond four, except when the effect of exposure is large. [2] In matched studies most efficient allocation of a fixed number of controls is usually one that sets the ratio of controls to cases to be approximately equal. [3] Therefore, choosing a ratio of 0.38 controls seems very inadequate. I am unable to understand how, age and sex matching with this ratio was conducted. Furthermore, would matching for just age and sex alone not have avoided all the confounding possible in this study? In addition, the use of the word proportion over prevalence would have been more appropriate.

References