On 2016 Mar 04, Andrea Margulis commented:

Diabetes progression and the consequent need for intensification of treatment occur in stages over time, with feedback loops and several influences including those from diabetes complications, comorbid conditions, and concomitant medications. The resulting web of prognosis, prescription, and disease progression over time creates methodological challenges for research on the effect of diabetes treatments [1,2]. Patients under different diabetes therapies may have very different baseline characteristics that may be only partially amenable to standard observational research techniques for confounding control. As patients pass through different stages of diabetes, treatment escalates or changes, and these updated therapeutic schemes may become indicators for a more severe stage of diabetes with a more guarded prognosis. In addition, each medication carries its own risks. In this situation, it becomes unclear how to validly assess exposure to earlier medications such as metformin, since continuing follow-up under a new therapeutic scheme mixes exposure effects and disease progression together, while stopping follow-up to isolate exposure to earlier therapeutics may create a bias similar to the differential loss to follow-up than can affect randomised controlled trials (informative censoring).

Challenges increase in the context of a comparison between treated diabetic patients and non-diabetic patients, as Bannister and colleagues [3] did with regard to the assessment of all-cause mortality. The authors report a 15% survival benefit in patients with diabetes on metformin monotherapy relative to patients without diabetes who were not on metformin, even though patients with diabetes were sicker at baseline. Results were consistent across subgroups, and a similar benefit was not observed for sulfonylurea treatment. This study addresses an important question, as previous research on metformin has produced promising findings, such as the ability to prolong lifespan and improve healthful aging in mice [4].

Although the results from Bannister et al. [3] are intriguing, we wonder whether study design features might have created a cohort of patients with diabetes that selectively retained healthier individuals, while the same was not true for the cohort without diabetes to which they were compared. To identify the effect of metformin or sulfonylurea monotherapy, follow-up of treated patients with diabetes was censored after any modification to the initial treatment. Thus, as patients with diabetes progressed to the point they needed treatment intensification, they left the cohort. In contrast, the patients without diabetes would not have the chance to be excluded from the cohort in a similar manner with any progression from a starting point of good health. This censoring is tied to the mortality outcome since if a patient with diabetes dies, the matched comparator patient cannot leave the cohort until he or she dies. On the other hand, if the comparator without diabetes dies first, the matched patient with diabetes can still leave the cohort upon treatment modification. This creates a different opportunity for mortality to be observed in the two cohorts.

Furthermore, the requirement of patients with diabetes to stay on glucose-lowering therapy for 180 days to be eligible (while comparators without diabetes are only required to survive for the 180 days) may have created an additional opportunity to selectively retain healthier individuals, i.e., adherent patients. Good adherence by itself has been shown to reduce mortality [5]. In subgroup analyses where patients with and without diabetes received cardiovascular prophylaxis (making them more similar in terms of adherence) the survival benefit decreases or disappears.

While the authors have posed an interesting and relevant research question, it is accompanied by many difficulties from a methodological perspective. It may be worth considering an alternative comparison cohort comprised of non-diabetic persons with treatment for some other chronic condition, for which an analogous run-in period and censoring upon treatment intensification could be applied. Such an approach would enhance comparability of the cohorts at baseline and provide similar opportunity for follow-up to stop when patients in either cohort change treatment; all of which would mitigate the healthy adherer effect in the comparison group. The validity of this approach would depend on the possibility for both arms to follow similar patterns of treatment modification in relation to the risk of death. This proposed alternative approach has the drawback of answering a somewhat different research question, but one that may be more amenable to observational investigation.

Andrea V Margulis, MD, ScD ¹

Manel Pladevall, MD, PhD ¹

Nuria Riera-Guardia, PhD ¹

John Seeger, PharmD, DrPH ² ³

Elisabetta Patorno MD, DrPH ²

Cristina Varas-Lorenzo, MD, PhD ¹

¹ RTI Health Solutions

² Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham & Women’s Hospital/Harvard Medical School

³ Optum Epidemiology

REFERENCES

1 Patorno E, Patrick AR, Garry EM, et al. Observational studies of the association between glucose-lowering medications and cardiovascular outcomes: addressing methodological limitations. Diabetologia. 2014 Nov;57(11):2237-50.

2 Suissa S, Azoulay L. Metformin and the risk of cancer: time-related biases in observational studies. Diabetes Care. 2012 Dec;35(12):2665-73.

3 Bannister CA, Holden SE, Jenkins-Jones S, et al. Can people with type 2 diabetes live longer than those without? A comparison of mortality in people initiated with metformin or sulphonylurea monotherapy and matched, non-diabetic controls. Diabetes Obes Metab. 2014 Nov;16(11):1165-73.

4 Martin-Montalvo A, Mercken EM, Mitchell SJ, et al. Metformin improves healthspan and lifespan in mice. Nat Commun. 2013;4:2192.

5 Granger BB, Swedberg K, Ekman I, et al. Adherence to candesartan and placebo and outcomes in chronic heart failure in the CHARM programme: double-blind, randomised, controlled clinical trial. Lancet. 2005 Dec 10;366(9502):2005-11.

On 2016 Mar 04, Andrea Margulis commented:

Diabetes progression and the consequent need for intensification of treatment occur in stages over time, with feedback loops and several influences including those from diabetes complications, comorbid conditions, and concomitant medications. The resulting web of prognosis, prescription, and disease progression over time creates methodological challenges for research on the effect of diabetes treatments [1,2]. Patients under different diabetes therapies may have very different baseline characteristics that may be only partially amenable to standard observational research techniques for confounding control. As patients pass through different stages of diabetes, treatment escalates or changes, and these updated therapeutic schemes may become indicators for a more severe stage of diabetes with a more guarded prognosis. In addition, each medication carries its own risks. In this situation, it becomes unclear how to validly assess exposure to earlier medications such as metformin, since continuing follow-up under a new therapeutic scheme mixes exposure effects and disease progression together, while stopping follow-up to isolate exposure to earlier therapeutics may create a bias similar to the differential loss to follow-up than can affect randomised controlled trials (informative censoring).

Challenges increase in the context of a comparison between treated diabetic patients and non-diabetic patients, as Bannister and colleagues [3] did with regard to the assessment of all-cause mortality. The authors report a 15% survival benefit in patients with diabetes on metformin monotherapy relative to patients without diabetes who were not on metformin, even though patients with diabetes were sicker at baseline. Results were consistent across subgroups, and a similar benefit was not observed for sulfonylurea treatment. This study addresses an important question, as previous research on metformin has produced promising findings, such as the ability to prolong lifespan and improve healthful aging in mice [4].

Although the results from Bannister et al. [3] are intriguing, we wonder whether study design features might have created a cohort of patients with diabetes that selectively retained healthier individuals, while the same was not true for the cohort without diabetes to which they were compared. To identify the effect of metformin or sulfonylurea monotherapy, follow-up of treated patients with diabetes was censored after any modification to the initial treatment. Thus, as patients with diabetes progressed to the point they needed treatment intensification, they left the cohort. In contrast, the patients without diabetes would not have the chance to be excluded from the cohort in a similar manner with any progression from a starting point of good health. This censoring is tied to the mortality outcome since if a patient with diabetes dies, the matched comparator patient cannot leave the cohort until he or she dies. On the other hand, if the comparator without diabetes dies first, the matched patient with diabetes can still leave the cohort upon treatment modification. This creates a different opportunity for mortality to be observed in the two cohorts.

Furthermore, the requirement of patients with diabetes to stay on glucose-lowering therapy for 180 days to be eligible (while comparators without diabetes are only required to survive for the 180 days) may have created an additional opportunity to selectively retain healthier individuals, i.e., adherent patients. Good adherence by itself has been shown to reduce mortality [5]. In subgroup analyses where patients with and without diabetes received cardiovascular prophylaxis (making them more similar in terms of adherence) the survival benefit decreases or disappears.

While the authors have posed an interesting and relevant research question, it is accompanied by many difficulties from a methodological perspective. It may be worth considering an alternative comparison cohort comprised of non-diabetic persons with treatment for some other chronic condition, for which an analogous run-in period and censoring upon treatment intensification could be applied. Such an approach would enhance comparability of the cohorts at baseline and provide similar opportunity for follow-up to stop when patients in either cohort change treatment; all of which would mitigate the healthy adherer effect in the comparison group. The validity of this approach would depend on the possibility for both arms to follow similar patterns of treatment modification in relation to the risk of death. This proposed alternative approach has the drawback of answering a somewhat different research question, but one that may be more amenable to observational investigation.

Andrea V Margulis, MD, ScD ¹

Manel Pladevall, MD, PhD ¹

Nuria Riera-Guardia, PhD ¹

John Seeger, PharmD, DrPH ² ³

Elisabetta Patorno MD, DrPH ²

Cristina Varas-Lorenzo, MD, PhD ¹

¹ RTI Health Solutions

² Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham & Women’s Hospital/Harvard Medical School

³ Optum Epidemiology

REFERENCES

1 Patorno E, Patrick AR, Garry EM, et al. Observational studies of the association between glucose-lowering medications and cardiovascular outcomes: addressing methodological limitations. Diabetologia. 2014 Nov;57(11):2237-50.

2 Suissa S, Azoulay L. Metformin and the risk of cancer: time-related biases in observational studies. Diabetes Care. 2012 Dec;35(12):2665-73.

3 Bannister CA, Holden SE, Jenkins-Jones S, et al. Can people with type 2 diabetes live longer than those without? A comparison of mortality in people initiated with metformin or sulphonylurea monotherapy and matched, non-diabetic controls. Diabetes Obes Metab. 2014 Nov;16(11):1165-73.

4 Martin-Montalvo A, Mercken EM, Mitchell SJ, et al. Metformin improves healthspan and lifespan in mice. Nat Commun. 2013;4:2192.

5 Granger BB, Swedberg K, Ekman I, et al. Adherence to candesartan and placebo and outcomes in chronic heart failure in the CHARM programme: double-blind, randomised, controlled clinical trial. Lancet. 2005 Dec 10;366(9502):2005-11.