Memorandum

30% of patients with type 2 diabetes initially decline insulin therapy; Average delay of >two years – November 24, 2017

Diabetic Medicine recently published a study from Brigham and Women’s Hospital showing that 30% of type 2 diabetes patients who were recommended insulin therapy declined it. Among this group, only 38% of people eventually started taking insulin within the study period (2000-2014), with a mean time to initiation of 790 days (2.2 years). Using records from 2000-2014, 3,295 insulin-naïve adults with type 2 diabetes who were recommended insulin therapy were identified. From this group, 984 (30%) initially declined insulin therapy. Those with a higher A1c (>9%) were more likely to decline insulin (34%) vs. those with an A1c <9% (21%) – in other words, patients who could benefit the most from insulin seemed to be warier of it and to oppose it. Of the 984 who declined insulin, 374 eventually started insulin therapy (38%), while 62% continued without insulin. While clinical inertia comes to mind when considering barriers to treatment intensification with insulin (a complicated drug when it comes to titration/dosing), this study raises the important point that patients also show resistance to insulin initiation (although it may also be associated, of course, with provider impressions). As we’ve heard countless times at diabetes meetings, insulin continues to be stigmatized as a “last resort” and sign of failure. More work needs to be done to promote an understanding of diabetes as a progressive disease and of insulin as a tool for better diabetes care. HCPs should also be aware of strategies to make patients more comfortable with insulin dosing and injection, and of course there’s a big role for digital health/dose titration apps to play in supporting patients on insulin therapy. This study validated and then used a natural language processing algorithm that was able to identify the refusal of insulin therapy from text of physician notes. This algorithm was designed using Canary, a public, natural language processing platform, and was validated using 1,501 random notes from PCPs in a large medical center. The algorithm achieved sensitivity of 100% in identifying insulin decline (95% CI: 82.4-100).

 

-- by Ann Carracher, Payal Marathe, and Kelly Close