Swedish Prospective, Noninterventional Study Design
The design and results of the Swedish noninterventional study have been published previously . In brief, a one-arm, prospective, observational, proof-of-concept study evaluated the introduction of a near-field communication (NFC)-enabled smart insulin pen (NovoPen® 6; Novo Nordisk A/S, Bagsværd, Denmark) in 94 adults with T1D receiving basal–bolus insulin therapy and using CGM, from 12 diabetes clinics in Sweden . Participants were included in the study at the discretion of their healthcare team and received a smart insulin pen for basal and/or bolus insulin injections. Downloadable long-term injection data were blinded to participants during the baseline period (baseline until visit 1), but CGM data could be viewed, and the dose of the last injection was displayed on the smart insulin pen . At visit 1, the first set of injection data was downloaded at the participant’s diabetes clinic, and subsequent follow-up visits were scheduled according to standard clinical practice. Smart insulin pen and CGM data were downloaded at each visit (CGM data were also uploaded between visits), and discussed and acted upon by the participant and healthcare team .
Glycemic summary measures and the number of missed bolus dose injections were compared between the blinded baseline period and the follow-up period . The follow-up period was defined as any point after the fifth visit, to allow for adequate discussion of available smart insulin pen data with the healthcare team . All 94 participants were included in the TIR analysis, while the 81 participants who were using the smart insulin pen for bolus insulin dosing were included in the bolus dose analyses and missed bolus dose analysis . TIR was defined as the time spent with sensor glucose within the acceptable range (3.9–10.0 mmol/L [70–180 mg/dL]) . Time spent in hypoglycemia (divided into level 1, 3.0 to < 3.9 mmol/L [54 to < 70 mg/dL]) and level 2, < 3.0 mmol/L [< 54 mg/dL]) , total daily insulin dose, mean glucose level, and the coefficient of variation were additional outcomes .
Cost-Effectiveness Model Overview
A cost-effectiveness analysis was performed by projecting costs (2018 Swedish krona, SEK) and clinical outcomes over patients’ lifetimes following the introduction of a smart insulin pen in a Swedish T1D population. This approach aims to capture the development of diabetes-related complications and their impact on projected costs, life expectancy, and quality of life, in alignment with guidelines on the assessment of the cost-effectiveness of diabetes interventions . The analysis was performed using the IQVIA CORE Diabetes Model (IQVIA, Durham, NC), a validated computer-simulation model of diabetes . The model projected the development of diabetes-related complications, mortality, hypoglycemia over a 60-year time horizon (i.e., patients’ remaining lifetimes) based on risk factors such as glycated hemoglobin (HbA1c), blood pressure, lipids, and body mass index. First-order stochastic uncertainty was captured via a Monte Carlo approach. Modeled outcomes included direct medical costs, indirect costs, life expectancy, quality-adjusted life expectancy, and the cumulative incidence and time to onset of diabetes-related complications in each simulation arm. It was assumed that patients continued to receive their set intervention (smart insulin pen in addition to standard care versus standard care) for the duration of their lifetime. Future costs and clinical outcomes were discounted at a rate of 3% per annum, in alignment with guidelines for economic evaluations in Sweden . This article does not contain any new studies with human participants or animals performed by any of the authors.
Model inputs for clinical outcomes were informed by the Swedish prospective, noninterventional study . In the study, smart insulin pens were associated with an additional 1.89 h per day TIR compared with the baseline measurement (without a smart insulin pen; 9.19 h, over a median follow-up of 7 months), and daily bolus insulin dose increased from 25.1 units [U] to 32.1 U from baseline to visit 5 . There are currently no published risk equations allowing TIR to be directly linked to incidence of diabetes-related complications and, therefore, a conversion from TIR to HbA1c was made, based on a previously identified linear relationship between the two parameters . A 10% change in TIR was considered equivalent to a change in HbA1c of 0.8% (9 mmol/mol) . TIR as a percentage (baseline, 41.4%; end of trial, 49.9%; change in TIR, + 8.5%) was converted to HbA1c using the following regression equation (HbA1c = [TIR − 155.4]/− 12.762) . This conversion allowed long-term outcomes to be modeled on the basis of published risk equations that use HbA1c as a risk factor for developing complications. These calculated HbA1c values at baseline (8.93%) and end of trial (8.27%) resulted in a HbA1c treatment effect of − 0.67% (− 7 mmol/mol), which was applied in the smart insulin pens arm, with no HbA1c treatment effect applied in the standard care arm. The HbA1c treatment effect was assumed to be constant in both arms of the analysis, with the difference maintained for the duration of the analysis. Hypoglycemic event rates were based on the study period of the Swedish prospective, noninterventional study for the smart insulin pens arm, and the baseline value for the standard care arm (3287.25 and 6574.50 events per 100 patient-years, respectively). For modeling purposes, hypoglycemia was defined as at least 15 min of below 54 mg/dL (< 3.0 mmol/L) in CGM data , and all hypoglycemic events were categorized as nonsevere. Mortality as a result of diabetes-related complications and background mortality based on Sweden-specific life tables  were applied.
All baseline characteristics were taken from people with T1D enrolled in the Swedish National Diabetes Register , other than baseline HbA1c (calculated from baseline TIR from the Swedish prospective, noninterventional study) , number of cigarettes smoked per day, and mean weekly alcohol consumption (both assumed to be the same as the Swedish general population) [25, 26] (Table 1). The model simulated the lifetime progression of diabetes in a cohort of 1000 hypothetical patients and repeated the process 1000 times for each simulation arm. This simulation generated mean and standard deviation values of clinical effectiveness and costs.
Quality-adjusted life expectancy was assessed using the CORE Default Method, which involves taking the lowest state utility associated with existing complications and adding event utilities for any events that occur in that year (Supplementary Table S1), to create annual utility scores for each simulated patient. Utilities associated with diabetes and disease-related complications were extracted from published sources (Supplementary Table S1).
Costs were estimated from a Swedish societal perspective to capture all direct medical costs (pharmacy costs, costs associated with diabetes-related complications, and concomitant patient management costs) and indirect costs as a result of lost productivity. Insulin doses applied in this model were based on the baseline values of patients enrolled in the Swedish prospective, noninterventional study for the standard care arm, and based on end-of-trial data for the smart insulin pens arm (daily bolus dose increased from 25.1 U at baseline to 32.1 U at visit 5) . Resource use and costs relating to insulin delivery were based on prices in Sweden, and are shown in Supplementary Table S2. Smart insulin pens and conventional durable insulin pens (without smart technology) have the same cost in Sweden (SEK 536.05) ; these costs were excluded from the analysis for simplicity owing to the relatively low costs and the durability (up to 5 years) of the insulin pens (Supplementary Table S2). Resource use relating to patient management was assumed to be the same as the general population of people with T1D in Sweden and is detailed along with patient management costs in Supplementary Table S3.
The cost of diabetes-related complications in the year of the event and for annual follow-ups (applied in each year of the simulation subsequent to the event) were identified through a literature review, and costs were inflated to 2018 values (Supplementary Table S4). Indirect costs of diabetes-related complications were considered in both arms of the analysis. Working age was considered to be from 20 to 65 years of age, with an average annual salary of SEK 438,000 for men and SEK 391,200 for women , and the working year considered to be 250 days. Days off work estimates were taken from an analysis of Danish registry data  or a study on the annual cost of hypoglycemia in Sweden , or a conservative assumption was used when no estimates could be identified (Supplementary Table S5). Cost results are presented in SEK, with a conversion to euros (EUR) in the Supplementary Appendix using a SEK 0.091 exchange rate, correct as of 27 March 2020.
As the extrapolation of clinical results by modeling long-term outcomes is associated with a level of uncertainty, sensitivity analyses were performed to assess the robustness of the findings and identify key drivers of modeled outcomes. Sensitivity analyses were performed by varying model parameters. The influence of time horizon on the outcomes projected by the model was investigated by running analyses over 3, 5, and 10 years. The base-case analysis used the baseline HbA1c from the Swedish prospective, noninterventional study, and sensitivity analyses were performed with variation in this input parameter. The calculated change in HbA1c (with the smart insulin pen relative to standard care) was varied to investigate uncertainty around the impact of smart insulin pens on glycemic control and the relationship between TIR and HbA1c. Further sensitivity analyses were conducted removing treatment effects in terms of HbA1c and hypoglycemic event rates, and with treatment switching from standard care to the smart insulin pen.