On 2015 Feb 11, Michael E Miller commented:

The subgroup analyses reported in our paper were in fact not affected by imbalance between groups. Our finding that “Compared to standard therapy, ACCORD’s intensive glycemia strategy resulted in a higher incidence of cardiovascular mortality in the younger participants but not in older participants (p=0.03 for interaction)” was not a comparison of cardiovascular mortality in younger versus older participants in the same treatment. It compared the effect of the allocated intervention within younger (HR=1.71) versus the effect within older participants (HR=0.97). Thus, among younger participants, the intensive strategy resulted in a higher CV mortality rate compared to the standard strategy; whereas, this treatment group effect was not observed among the older participants. With approximately 3400 participants in both intensive and standard glycemia groups within younger participants and 1700 participants in each group among older participants, baseline characteristics of the intensive and standard groups within age groups were well-balanced. The suggestion that “differences in the baseline characteristics, not the glycemic control in the intensive group might have been responsible for increased mortality in younger participants” is therefore incorrect. Indeed, mortality was not greater in the younger group (see Figure 1). We wrote: “As expected, the older subgroup had higher absolute event rates for all outcomes considered within both treatment arms.” Were we to statistically compare incidence rates in younger versus older participants within the same treatment (which our paper did not do), then clearly many health related characteristics of younger and older participants would be different (as we reported in Supplementary Table 1 of the online appendix) and this would contribute to any differences within treatment groups between age groups.

On 2015 Feb 07, Jayaprakash Sahoo commented:

We read with interest the article by Miller et al in which they have analyzed the impact of baseline age on the effect of intensive blood glucose lowering in the ACCORD trial(1). One of the conclusions of this analysis is that intensive glycemic strategy results in a higher incidence of cardiovascular mortality in the younger but not in older participants. This odd conclusion has to be subjected to scrutiny.

There is a concept that the quality of output is determined by the quality of the input and this applies to statistical analysis also. The basic requirement for comparison of two groups (input data) prior to any analysis is matching of their baseline characteristics. Any conclusion (output data) reached as a result of an analysis without matching is bound to be associated with bias. The baseline characteristics of the intensive and standard therapy arms of the original ACCORD trial were perfectly matched (2). So, the increased mortality in the intensive arm might have been the effect of glycemic control per se in that study. However, the post hoc analysis of the original ACCORD data comparing the adverse events, cardiovascular disease and mortality in older versus younger adults has discrepant baseline characteristics like body mass index (BMI), waist circumference (WC), glycated hemoglobin (HbA1c), blood pressure (BP), and lipid profile. It is probably because this was not the primary objective of the original study. The older participants had a favorable risk factor profile of significantly lower BMI, WC, HbA1c, diastolic BP, LDL cholesterol, triglycerides and high HDL cholesterol as compared to younger participants. The differences in the baseline characteristics, not the glycemic control in the intensive group might have been responsible for increased mortality in younger participants. The authors have discussed role of hypoglycemia and rate of decline in HbA1C as possible causes of increased mortality, but they could not reach any conclusion (3-4). They have not touched upon the impact of differences in baseline characteristics on mortality. In addition, the contributions from blood pressure and lipid control towards the end points leading to inferences in this study has to be taken into account(5). So, the conclusions drawn from analysis of retrospective data, where two groups compared had different baseline characteristics should be taken with caution.

References:

1. Miller ME, Williamson JD, Gerstein HC, et al. Effects of randomization to intensive glucose control on adverse events, cardiovascular disease, and mortality in older versus younger adults in the ACCORD trial. Diabetes Care 2014; 37:634-643.

2. Gerstein HC, Miller ME, Byington RP, et al.; Action to Control Cardiovascular Risk in Diabetes Study Group. Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med 2008; 358:2545–2559.

3. Bonds DE, Miller ME, Bergenstal RM, et al. The association between symptomatic, severe hypoglycaemia and mortality in type 2 diabetes: retrospective epidemiological analysis of the ACCORD study. BMJ 2010; 340:b4909.

4. Riddle MC, Ambrosius WT, Brillon DJ, et al.; Action to Control Cardiovascular Risk in Diabetes Investigators. Epidemiologic relationships between A1C and all-cause mortality during a median 3.4-year follow up of glycemic treatment in the ACCORD trial. Diabetes Care 2010; 33:983–990.

5. Margolis KL, O'Connor PJ, Morgan TM, et al. Outcomes of combined cardiovascularrisk factor management strategies in Type 2 diabetes: The ACCORD randomized trial. Diabetes Care. 2014 Mar 4.

On 2015 Feb 07, Jayaprakash Sahoo commented:

We read with interest the article by Miller et al in which they have analyzed the impact of baseline age on the effect of intensive blood glucose lowering in the ACCORD trial(1). One of the conclusions of this analysis is that intensive glycemic strategy results in a higher incidence of cardiovascular mortality in the younger but not in older participants. This odd conclusion has to be subjected to scrutiny.

There is a concept that the quality of output is determined by the quality of the input and this applies to statistical analysis also. The basic requirement for comparison of two groups (input data) prior to any analysis is matching of their baseline characteristics. Any conclusion (output data) reached as a result of an analysis without matching is bound to be associated with bias. The baseline characteristics of the intensive and standard therapy arms of the original ACCORD trial were perfectly matched (2). So, the increased mortality in the intensive arm might have been the effect of glycemic control per se in that study. However, the post hoc analysis of the original ACCORD data comparing the adverse events, cardiovascular disease and mortality in older versus younger adults has discrepant baseline characteristics like body mass index (BMI), waist circumference (WC), glycated hemoglobin (HbA1c), blood pressure (BP), and lipid profile. It is probably because this was not the primary objective of the original study. The older participants had a favorable risk factor profile of significantly lower BMI, WC, HbA1c, diastolic BP, LDL cholesterol, triglycerides and high HDL cholesterol as compared to younger participants. The differences in the baseline characteristics, not the glycemic control in the intensive group might have been responsible for increased mortality in younger participants. The authors have discussed role of hypoglycemia and rate of decline in HbA1C as possible causes of increased mortality, but they could not reach any conclusion (3-4). They have not touched upon the impact of differences in baseline characteristics on mortality. In addition, the contributions from blood pressure and lipid control towards the end points leading to inferences in this study has to be taken into account(5). So, the conclusions drawn from analysis of retrospective data, where two groups compared had different baseline characteristics should be taken with caution.

References:

1. Miller ME, Williamson JD, Gerstein HC, et al. Effects of randomization to intensive glucose control on adverse events, cardiovascular disease, and mortality in older versus younger adults in the ACCORD trial. Diabetes Care 2014; 37:634-643.

2. Gerstein HC, Miller ME, Byington RP, et al.; Action to Control Cardiovascular Risk in Diabetes Study Group. Effects of intensive glucose lowering in type 2 diabetes. N Engl J Med 2008; 358:2545–2559.

3. Bonds DE, Miller ME, Bergenstal RM, et al. The association between symptomatic, severe hypoglycaemia and mortality in type 2 diabetes: retrospective epidemiological analysis of the ACCORD study. BMJ 2010; 340:b4909.

4. Riddle MC, Ambrosius WT, Brillon DJ, et al.; Action to Control Cardiovascular Risk in Diabetes Investigators. Epidemiologic relationships between A1C and all-cause mortality during a median 3.4-year follow up of glycemic treatment in the ACCORD trial. Diabetes Care 2010; 33:983–990.

5. Margolis KL, O'Connor PJ, Morgan TM, et al. Outcomes of combined cardiovascularrisk factor management strategies in Type 2 diabetes: The ACCORD randomized trial. Diabetes Care. 2014 Mar 4.

On 2015 Feb 11, Michael E Miller commented:

The subgroup analyses reported in our paper were in fact not affected by imbalance between groups. Our finding that “Compared to standard therapy, ACCORD’s intensive glycemia strategy resulted in a higher incidence of cardiovascular mortality in the younger participants but not in older participants (p=0.03 for interaction)” was not a comparison of cardiovascular mortality in younger versus older participants in the same treatment. It compared the effect of the allocated intervention within younger (HR=1.71) versus the effect within older participants (HR=0.97). Thus, among younger participants, the intensive strategy resulted in a higher CV mortality rate compared to the standard strategy; whereas, this treatment group effect was not observed among the older participants. With approximately 3400 participants in both intensive and standard glycemia groups within younger participants and 1700 participants in each group among older participants, baseline characteristics of the intensive and standard groups within age groups were well-balanced. The suggestion that “differences in the baseline characteristics, not the glycemic control in the intensive group might have been responsible for increased mortality in younger participants” is therefore incorrect. Indeed, mortality was not greater in the younger group (see Figure 1). We wrote: “As expected, the older subgroup had higher absolute event rates for all outcomes considered within both treatment arms.” Were we to statistically compare incidence rates in younger versus older participants within the same treatment (which our paper did not do), then clearly many health related characteristics of younger and older participants would be different (as we reported in Supplementary Table 1 of the online appendix) and this would contribute to any differences within treatment groups between age groups.