Feasibility of Using a Smart Phone Application for Self-titration of Insulin on Glycemic Control in Patients With T2DM

Background:

Many patients with type 2 diabetes mellitus (T2DM) require insulin therapy after suboptimal glycemic control with oral antidiabetic agents. Initiating insulin therapy and adjusting dosage is vital in improving clinical outcomes among patients with diabetes. However, timely and appropriate dosage titration is challenging for many patients. Doctors also contend with many patients lacking the confidence to self-titrate insulin. Hence, reinforcing patients' self-empowerment is important in diabetes management. Developing tools that facilitate dosage adjustment of insulin may have substantial benefits to glycemic control.

Objectives:

The overall aim of this study is to investigate the feasibility of using a patient-centered smart phone application for insulin self-titration to improve glycemic control relative to usual care in patients with T2DM. The specific aim of the study is to determine the technical and clinical feasibility of a patient-centered smart phone application for insulin self-titration on glycemic control (HbA1c) of suboptimally controlled patients with T2DM who are starting insulin treatment.

Study Design:

This 24-week study is a randomized controlled trial that will involve participants recruited from Singapore General Hospital, a 1500-bed tertiary care hospital. The target population will consist of insulin naive patients with T2DM (30-70 years old) with suboptimal glycemic control (HbA1c ≥ 7.5%) while on oral antidiabetic agents.

Sample Size:

Eighty patients will be recruited in the study and randomized into either the intervention or control group in equal numbers (n=40) within strata defined by provider (i.e., within each of the five doctors). Patients will be nested within providers, and randomization within each provider will be performed in blocks of 4 subjects: 2 for the intervention and 2 for the control.

Subject Selection:

Eligible patients will be recruited over a 18-month period from the Singapore General Hospital. All patients newly initiated on insulin therapy will receive standard diabetes education from Diabetes Nurse Educators (DNEs). The research coordinator and doctors will work closely with the DNEs to identify eligible patients for this study. When a patient is eligible to participate in the study, the research coordinator will provide a study information brochure and explain the study design to the patient. Investigators will obtain an informed consent from patients prior to randomization.

Intervention:

Use of a smart phone application in self-titration of insulin

Primary Outcome Variable:

Difference in change in glycemic control (HbA1c) between the intervention and the control group

Procedures:

Recruited patients will receive once-daily injections of insulin detemir, a long-acting insulin, to be given at bedtime. The physicians are free to keep patients on all oral antidiabetic agents, except for thiazolidinediones. All recruited patients will receive free glucometers and test strips.

Statistical Analysis:

The primary analysis will be a comparison of mean reduction in HbA1c at 24 weeks post-randomization in the treatment and control groups in the context of a linear mixed-effects model with random effects for patients and doctors. The study is randomized, so baseline features and potential confounders should be approximately balanced between the treatment and control groups. However, the sample size is small (80 patients in total) so the mixed-effects model will be adjusted for covariates selected using forward stepwise variable selection with the AIC criterion. Potential covariates include age, gender, whether the patient themselves has a smart-phone, and treatment satisfaction.

The mixed-effects model allows all non-missing data to be used in the analysis without imputation. However, for the estimate of the difference in reduction in HbA1c to be unbiased, it is required that the distribution of the data missingness conditional on the variables in the model and the observed data does not depend on the values of the missing data, so that the data is missing-at-random.This is a common and a reasonable assumption. However, it is possible that the analysis of attrition will indicate that additional covariates need to be included in the mixed-effects model. With the aim of developing a tailored analysis plan for a definitive study, a few additional models for comparison of mean reduction in HbA1c will also be considered. For example, a linear mixed-effects model with random slope and intercept for each patient and doctor which incorporates the intermediate HbA1c measurements for each patient will also be fit. It is unknown a priori whether such a model will have reduced power because of patients attaining equilibrium titration at different times or if such a model will have increased power because it uses more observations albeit on the same group of patients. Depending on the rate of attrition and whether attrition is related to covariates, a propensity score for missingness may be developed and models fit which are adjusted for or stratified by this score.

The primary examination of efficacy will be through a linear mixed-effects model with random effects for each doctor and patient, under the assumption that the distribution of missingness conditional on the covariates and observed data does not depend on the missing data values. However, the investigators will examine several analyses for appropriateness including simple comparisons of the reductions in HbA1c between the treatment and control groups under last-observation-carried-forward, which is expected to be fairly conservative, and worst-observation-carried-forward, which is expected to be extremely conservative. What these methods lack in rigor may be offset by their simplicity.