A 54-year-old man comes to the office for follow-up of type 2 diabetes mellitus. The patient also has a history of hypertension, obesity, and chronic kidney disease (CKD). Medications include an antihyperglycemic agent and insulin. Blood pressure is 148/92 mm Hg and pulse is 81/min and regular. BMI is 34 kg/m². Physical examination is otherwise unremarkable. Laboratory results are as follows: Based on his serum creatinine level, his CKD has progressed, and the calculated glomerular filtration rate (eGFR) is now 42 mL/min/1.73 m², down from 55 mL/min/1.73 m² as measured 6 months ago. As stated in the ad, the proportion of subjects reaching A1c

Question

A 54-year-old man comes to the office for follow-up of type 2 diabetes mellitus. The patient also has a history of hypertension, obesity, and chronic kidney disease (CKD) . Medications include an antihyperglycemic agent and insulin. Blood pressure is 148/92 mm Hg and pulse is 81/min and regular. BMI is 34 kg/m². Physical examination is otherwise unremarkable. Laboratory results are as follows:

Based on his serum creatinine level, his CKD has progressed, and the calculated glomerular filtration rate (eGFR) is now 42 mL/min/1.73 m², down from 55 mL/min/1.73 m² as measured 6 months ago.
As stated in the ad, the proportion of subjects reaching A1c <7.0% was one of the prespecified secondary endpoints evaluated at week 26. Compared to the placebo group, a greater proportion of subjects in the 2 kanebaflozin groups achieved A1c <7.0%; however, researchers did not perform or report tests for the statistical significance of the differences in secondary endpoints among the 3 groups. Which of the following is the most likely reason that these statistical tests were not conducted?

Quizplus · Accepted Answer

Explanation: The hypothesis testing process requires that error rates (eg, type I and type II errors) be defined a priori. Type I error (alpha) is the probability of rejecting a null hypothesis when the null hypothesis is true. Conducting multiple independent hypothesis tests without proper adjustment to the alpha level increases the rate of type I error. This means that, when evaluating multiple secondary endpoints (eg, fasting plasma glucose, proportion of subjects reaching A1c <7.0%, changes in renal parameters), there is a higher probability of erroneously finding a statistically significant result with one of these endpoints (eg, higher likelihood of type I error) due to chance alone. This phenomenon is known as the multiplicity, or multiple testing, problem. In general, the rate for type I error will increase depending on the alpha level for individual tests and the number of independent tests. For example, the rate for type I error in a study attempting to evaluate 5 secondary endpoints would be ~23%, whereas the classically accepted value is usually set ≤5% (statistically significant p-value). Statistical techniques can be used to account for the multiple testing problem (eg, the alpha level or p-value is sometimes adjusted). (Choice A) Results showed comparable mean baseline A1c among the 3 groups (7.8%, 8.0%, and 8.1% for placebo, kanebaflozin 100 mg, and kanebaflozin 300 mg, respectively). This suggests adequate randomization in the study and should not affect differences in the proportion of subjects achieving A1c <7.0% at 26 weeks. (Choice B) The proportion of subjects achieving A1c <7.0% at 26 weeks in the 2 kanebaflozin groups (26.7% and 34.5% for kanebaflozin 100 mg and 300 mg, respectively) might be clinically significant when compared to the smaller proportion of such subjects in the placebo group (15.8%). However, clinical significance should not be misinterpreted as statistical significance and must not drive the decision of whether to conduct a statistical analysis in a study. (Choice D) Nonoverlapping confidence intervals indicate a statistically significant difference between compared groups. However, the opposite is not necessarily true. Although there appears to be an overlap in confidence intervals around the estimates of the proportion of subjects reaching A1c <7.0% in each of the groups (15.8%, 26.7%, and 34.5%), this overlap does not imply statistical significance. (Choice E) In contrast to the kanebaflozin groups, for which there were mean decreases in A1c from baseline at 26 weeks (−0.34% and −0.46%), there was a mean increase in the placebo group (+0.05%). However, this does not provide any information regarding the statistical significance of the differences among the groups. Educational objective: There is an increased risk for type I errors (ie, false positives) when multiple simultaneous hypotheses are tested at set p-values. Interpretation of findings in clinical trials when multiple outcome measures are used without adjustment of the significance level may result in spurious findings and invalid conclusions.

A 54-Year-Old Man Comes to the Office for Follow-Up of Type