Reasons for discontinuing insulin and factors associated with insulin discontinuation in patients with type 2 diabetes mellitus: a real-world evidence study

Associated Data

Supplementary Materials

Data Availability Statement
The datasets generated during and/or analyzed during the current study are not publicly available due to the institutional patient privacy policy but are available from the corresponding author upon a reasonable request.

Abstract

Background

Type 2 diabetes mellitus (T2DM) is a highly prevalent chronic disease with serious medical and economic burden [1–3]. The major focus of disease management in patients with T2DM is glycemic control, which might be initially achieved through lifestyle change and treatment with first-line metformin [4]. Due to the progressive nature of this chronic disease, many patients with T2DM eventually will require treatment intensification with other medications, including insulin [5]. Lower incidence of diabetes complications and lower severity of diabetes as well as reduced patient-borne costs have been reported for patients with high adherence to insulin therapy [6, 7]. Despite American Diabetes Association (ADA) guidelines and compelling evidence regarding the benefits of insulin therapy, insulin is underutilized in patients with T2DM. Especially alarming is the number of patients who stop insulin therapy in spite of poorly controlled hyperglycemia [8, 9].

One important aspect of cessation of insulin therapy that has not been well studied is discontinuation of insulin therapy by healthcare providers. While non-adherence to insulin is driven primarily by the patient’s decision not to take the medication, discontinuation of insulin therapy countenanced by the patient’s healthcare provider is a modification of the treatment regimen that may have been prompted by a change in clinical circumstances. However, why insulin therapy is being discontinued is not completely understood. Important aspects of discontinuation of insulin therapy include both the reasons for discontinuation and treatment characteristics that are associated with higher incidence of insulin discontinuation. One reason for this knowledge gap is that most previous analyses of discontinuation of insulin therapy (a.k.a. insulin non-persistence) were restricted to administrative/claims data [10–13]. These data sources can identify patients who stopped taking their medication, but they contain little information about the reasons for discontinuation, and may not be able to differentiate between patient-driven non-persistence and insulin discontinuation recommended by a healthcare provider. Other studies used surveys or interviews to identify the factors and reasons associated with medication non-adherence or discontinuation [14, 15]. However, most of these have focused specifically on insulin non-adherence. Furthermore, the quality and validity of data obtained through these approaches may be affected by selection bias, recall bias and response rates [16, 17]. Finally, the cost, time and other resources involved in interviews may be prohibitive.

Electronic medical records (EMR) present a unique opportunity to evaluate both reasons and factors associated with insulin discontinuation. Many EMR systems allow clinicians to electronically record pertinent clinical information, including reasons for discontinuing a medication. Furthermore, EMR systems also include a variety of data that may influence insulin discontinuation decisions but is not available in claims data, such as vital signs / body mass and laboratory test measurements. EMR systems capture patient information using both structured data (based on controlled vocabularies) and unstructured narrative text. We therefore leveraged analysis of structured and unstructured EMR data to conduct this study to identify reasons for discontinuation of insulin and determine factors associated with insulin discontinuation in patients with T2DM.

Methods

Results

Study population

We identified 8744 patients with diabetes mellitus treated with insulin in a primary care or endocrinology clinic affiliated with one of the study institutions. After 1735 patients were removed from the analysis based on the inclusion and exclusion criteria (Table 1), 7009 patients were included in the study. Insulin was discontinued by 2957 (42.2%) study patients. Patients who discontinued insulin lost 0.10% of body weight while patients who did not discontinue insulin gained 0.90% of body weight over the study period (P <  0.001). Most (2509 / 84.8%) of insulin discontinuations were identified based on provider orders to stop the medication recorded in the EMR. Patients who had insulin discontinued had a greater decrease in HbA1c over the study period compared to patients who did not (Fig. 1). Majority (71.7%) of patients who discontinued insulin had at least one documented reason for discontinuation; demographic characteristics and comorbidities of patients who had discontinued insulin and those who had reasons for insulin discontinuation documented were similar to the overall study population (Table 2 and Supplemental Table 1).

Table 1

Patient flow chart

Criterion	Number of patients in the population after criterion applied
Patients who fulfill inclusion criteria	8744
No evidence of secondary diabetes	7168
No evidence of chronic pancreatitis or pancreatectomy	7072
No evidence of neonatal diabetes	7072
No urine ketone strips prescriptions	7071
No C-peptide < 0.2 pmol / L	7055
No anti-GAD or anti-islet antibodies	7009

Open in a separate window

Please note that this image might contain sensitive content. Click to unblur.

 

Open in a separate window

Fig. 1

Insulin Discontinuation and Glycemic Control. The figure represents the patients for whom HbA1c measurements were available (1588 out of 2957 patients who discontinued and 2100 out of 4052 patients who did not discontinue insulin therapy at baseline; 2584 out of 2957 patients who discontinued and 3547 out of 4052 patients who did not discontinue insulin therapy at end-of-study). Circles represent mean values. Wisps indicate standard error. Paired t-test was used to determine statistical significance (only for the 3580 patients who had both baseline and end-of-study HbA1c measurements available)

Table 2

Patient characteristics part I

Please note that this image might contain sensitive content. Click to unblur.

Open in a separate window

Cells where the metric is not applicable to the variable are asterisks

-: no missing data for this variable

BMI Body mass index, CCI Charlson Comorbidity Index, DBP Diastolic blood pressure, eGFR Estimated glomerular filtration rate, HbA1C Hemoglobin A1c, HTN Hypertension, LDL Low density lipoprotein cholesterol, PVD Peripheral vascular disease, SBP Systolic blood pressure, SD Standard deviation

Discussion

This population-based analysis leveraging EMR data of over 7000 patients with T2DM showed that many – over a third – discontinued insulin therapy. Among the patients with documented reasons for discontinuation of insulin, the most common reasons were improving blood glucose control, achieved weight loss and initiation of non-insulin diabetes medications. This is the first study, to our knowledge, that leveraged both structured and unstructured EMR data to understand insulin discontinuation.

The most common reason for insulin discontinuation was good blood glucose control. Second (achieved weight loss) and third (non-insulin diabetes medication started) most common reasons for insulin discontinuation were likely related to blood glucose control as well: achieved weight loss leads to a decrease in insulin resistance and ultimately to lower blood glucose levels and insulin requirements [19], while initiation of additional non-insulin medications may achieve improved blood glucose control on its own, obviating the need for insulin. Consistent with this explanation, these three reasons were commonly reported together. A typical example was a middle-aged patient whose diabetes was diagnosed when he was admitted to the hospital with a severe infection (Fournier gangrene). The patient was initially started on basal-bolus regimen. However, as his infection resolved and he improved his diet and lost weight, his blood glucose levels decreased and he was able to discontinue prandial insulin (continuing on a lower dose of basal insulin). At the population level, the finding of a greater decrease in HbA1c among patients whose insulin was discontinued was also consistent with this explanation. Adverse reactions to insulin, such as hypoglycemia, played a smaller role. In a number of cases hypoglycemia was reported together with good blood glucose control and may have been an unintended consequence of the overall lowering of blood glucose levels that ultimately enabled insulin discontinuation. Financial difficulties were reported as a reason for insulin discontinuation for < 10% of patients. This may reflect the fact that Massachusetts, where the study was conducted, has very high penetration of health insurance and this finding may not generalize to the rest of the U.S.

There were several patient and provider characteristics that were strongly associated with insulin discontinuation. The strongest factor for discontinuation of insulin therapy was treatment by an endocrinologist. This may have been due to more time and resources available to endocrinologists to provide lifestyle counseling and / or to their early adoption (and thus greater propensity to prescribe) novel non-insulin diabetes medications. Either of these approaches could lead to improvement in glycemic control and discontinuation of insulin. Female providers were also more likely to discontinue insulin. This could have been a residual effect of specialty – most endocrinologists at Partners HealthCare are women. Patients were more likely to have insulin discontinued if they were treated with a basal-bolus regimen. This could reflect the fact that more complicated insulin regimens can be more challenging to manage, and that could ultimately prompt insulin discontinuation. Patients whose BMI had risen after study entry were less likely to discontinue insulin. This finding could have been confounded: insulin therapy results in weight gain and therefore weight gain could have reflected the fact that treatment with insulin was continued.

Consistent with the objective of the investigation to analyze a distinct clinical phenomenon of insulin discontinuation, factors associated with and reasons for insulin discontinuation identified in our study differed significantly from those previously identified for insulin non-adherence. For example, while previously published studies reported strong associations of age and gender with non-adherence [20–22], neither was related to insulin discontinuation in our findings. Similarly, practical and logistical barriers commonly cited by patients as the reasons for insulin non-adherence [20, 21, 23] were not identified as the reasons for insulin discontinuation in our study.

This analysis of insulin discontinuation by providers was enabled by EMR data. Both claims and EMR data have their strengths and weaknesses [24]. One important distinction of EMR data that was particularly pertinent to the present study is that it contains information on prescriber orders. Claims data allows analysis of whether the patient is taking the medication or not (based on the assumption that patients are unlikely to purchase medications, often incurring copayments, if they are not taking them). However, it is impossible to determine from the claims data whether cessation of medication consumption by the patient was recommended by their healthcare provider. On the other hand, EMR data includes information on prescriber medication orders to both initiate and to discontinue medications. In particular, in our dataset almost all (nearly 85%) of insulin discontinuations were identified based on EMR provider orders to stop the medication. This allowed the present analysis to focus on a previously unexplored aspect of diabetes therapy – insulin discontinuation by healthcare providers.

Most reasons for insulin discontinuation were recorded in narrative provider notes and required NLP technology for their identification. This finding illustrates the opportunities accorded by the increasing availability of multiple streams of EMR data, including narrative electronic documents, combined with the advancements in artificial intelligence technologies that allow population-scale analyses of these data. In particular, natural language processing has shown increasing promise in allowing quantification of previously poorly understood clinical phenomena and their effects on patient outcomes [25–27]. The present study continues to demonstrate the power of this approach by leveraging natural language processing provide initial data on discontinuation of insulin therapy by healthcare providers.

This analysis had a number of strengths. It included a large sample of patients, the majority of whom were treated in a primary care setting, just as most patients with diabetes in the U.S. are. While relatively few patients had reasons for insulin discontinuation recorded in the traditionally used “structured” EMR data (medication lists, adverse medication reaction lists, etc.), this project drew on a novel information source – narrative electronic provider notes. It leveraged validated natural language processing technology to identify reasons for insulin discontinuation from a much larger number of patients than would have been possible either using structured EMR data alone (due to missing data) or manual record review (due to the time and costs involved in reviewing records of thousands of patients). This project therefore represents an advance in clinical research made possible by modern technology.

The findings of this project have to be interpreted in light of its limitations. The analysis was limited to a single integrated healthcare delivery system in Eastern Massachusetts; therefore, the results may not be generalizable to the rest of the U.S. Patients were only followed for 12 months; hence the findings may not be applicable to insulin discontinuation later in the treatment course. Information on the patients’ duration of diabetes was not available for analysis. Some of the patients who had discontinuation of insulin therapy established based on lack of insulin prescriptions for > 12 months may not have, in fact, discontinued insulin; therefore, the rate of insulin therapy discontinuation could have been overestimated. While the accuracy of the natural language processing tools was high, it was not perfect. Consequently, some insulin discontinuation reasons could have been missed and others misidentified. Furthermore, some insulin discontinuation reasons may not have been documented in the EMR at all. Some of the episodes of insulin discontinuation analyzed in the study, while recorded by healthcare providers, may have reflected patient-driven non-adherence. However, based on our clinical experience, patients who are not adherent to their medication that their provider feels is indicated for them seldom have it discontinued in their EMR record – consistent with the finding that patient preference was not commonly cited as a reason for insulin therapy discontinuation in the study data. There was a large amount of missing laboratory data; however, analyses that did not include variables with large amounts of missing data and analyses that imputed missing data had very similar results. Finally, as an observational study, the analysis of factors associated with insulin discontinuation could not identify causal relationships but only associations.

Conclusions

In summary, this study has confirmed that discontinuation of insulin therapy countenanced by healthcare providers is a distinct clinical phenomenon whose risk factors and reasons are different from the more widely studied medication non-adherence. While many patients discontinued insulin therapy, in most cases discontinuation appeared appropriate. Frequently it was made possible by patients achieving blood glucose control by alternative means, either lifestyle changes, additional non-insulin diabetes medications, or both. Adverse reactions to insulin (e.g. hypoglycemia or weight gain) and financial difficulties played a smaller role than blood glucose control (in an area with a high penetration of health insurance). Many of the factors associated with insulin discontinuation identified in the study (e.g. treatment by an endocrinologist, who may be more likely to use new non-insulin diabetes medications, or a basal-bolus insulin regimen that can be more challenging to manage) were consistent with the overall findings.

Supplementary Information

Acknowledgments

Abbreviations

Authors’ contributions

All authors confirm that they meet the International Committee of Medical Journals Editors uniform requirements for authorship. All authors have contributed to the design of the review, its drafting and/or revision, and have approved the final submitted version.

Funding

This study was funded by Eli Lilly and Company.

Availability of data and materials

The datasets generated during and/or analyzed during the current study are not publicly available due to the institutional patient privacy policy but are available from the corresponding author upon a reasonable request.

Ethics approval and consent to participate

This study was approved by the Partners HealthCare institutional review board, and the requirement for informed consent was waived.

Consent for publication

Not applicable (the manuscript does not contain any individual person’s data in any form).

Competing interests

Alexander Turchin declares research funding from Astra Zeneca, Edwards, Eli Lilly and Company, Novo Nordisk, Pfizer and Sanofi; consulting fees from Proteomics International; and equity in Brio Systems. Jianmin Wu, Zhenxiang Zhao, Ginger Haynes, Xuanyao He and Ayad Ali are employees and stockholders of Eli Lilly and Company. Xiaomei Peng is a former employee of Eli Lilly and Company.

The rest of the authors have no potential conflicts of interest to disclose, including relevant financial interests, activities, relationships, and affiliations with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript.

Footnotes

References