Research Design and Methods
Study Design
The design of, study population, and data collection for the MOSAIc study have been described extensively. In brief, MOSAIc is a 2-year, longitudinal observational study that follows patients' real world diabetes care and health outcomes. MOSAIc involves no additional treatments, visits, or laboratory collections beyond those occurring within the course of normal care. Data collection occurs during an initial baseline visit and during four subsequent prospective visit windows (within ±3 months) at 6, 12, 18, and 24 months. At present, only baseline data are available for all patients.
Study Population
MOSAIc has enrolled 4,519 patients from primary care or specialty practice sites across 18 countries [United Arab Emirates (UAE), Argentina, Brazil, Canada, China, Germany, India, Israel, Italy, Japan, Mexico, Russia, Saudi Arabia, South Korea, Spain, Turkey, United Kingdom (UK), United States including Puerto Rico (US)]. These participating countries were chosen based on geographic region, population aged 20–79, and the T2DM prevalence in each region and country. These countries have heterogeneous levels of economic development, industrialization, and healthcare accessibility and represent 5 global regions: Asia (China, India, Japan, South Korea), Europe (Germany, Italy, Russia, Spain, UK), North America (Canada, US), Middle East/North Africa (Israel, Saudi Arabia, Turkey, UAE), South/Central America (Argentina, Brazil, Mexico). Patients were recruited from two types of clinics, primary care practices and diabetes specialty clinics, to be representative of each participating country's diabetic patient population. In addition, we included sites in both rural and urban locations and academic and non-academic settings. Clinicians treating enrolled patients have also contributed demographic data and patient-specific treatment goals at baseline. MOSAIc-eligible patients were 1) age ≥18; 2) taking any commercially-available insulin therapy other than intensive basal-bolus insulin therapy (i.e., basal + 3 prandial injections) from any manufacturer for ≥ 3 months with or without any combination of approved non-insulin antidiabetic medications; 3) were not simultaneously participating in a study that includes an investigational drug or procedure; and 4) were proficient in the country's primary language. All enrolled patients gave documented consent. Due to inconsistent and poor data quality, 2 practice sites have been closed since the study began. Data from these sites were excluded from analyses in the present manuscript (Fig. 1). Institutional Review Board approvals were obtained in all countries. The Institutional Review Board of the Brigham and Women's Hospital, the data coordinating center for the entire MOSAIc study, deemed the study's analytic plan as exempt from review. A list of the ethics review boards for all countries is provided in Additional file 1 http://www.biomedcentral.com/1472-6823/15/46/additional.
(Enlarge Image)
Figure 1.
CONSORT diagram for MOSAIc baseline analytic population
Baseline Data Collection
Each patient's available type 2 diabetes clinical history, including diagnosis date, treatment and complications, and medication history, was assessed retrospectively from the practice site's medical record. Type 2 diabetes-related health care resource utilization (physician visits, hospitalizations, and auxiliary provider visits: diabetes educators, ophthalmologists, podiatrists, cardiologists, dietitians, and nephrologists), most recent recorded laboratory, biometric, and vital sign values, and other comorbidities were also assessed, but limited to the period 6 months prior to the baseline visit. Patients' type 2 diabetes medication regimen at the time of the baseline visit, including medication name, dose, frequency of use, method of administration (oral, syringe, pen, or pump) were collected. Finally, each patient's physician reported an HbA1c goal for him/her.
Extensive information on patients' diabetes—and insulin-related knowledge, attitudes, and behaviors; hypoglycemia and fasting; general health behaviors; patient-provider relationships; and perceived physical and psychological well being were collected at baseline using self-report questionnaires. In this paper, we focus on 3 of these validated surveys. The Diabetes Knowledge Test examines patients' understanding of their disease, such as how to manage insulin use and how to treat hypoglycemia, with a summary score ranging from 0 (no questions correct) to 9 (all correct). The 17-item Diabetes Distress Scale asks patients to indicate to what degree aspects of their type 2 diabetes treatment and care are of concern, using a 6-point Likert scale ranging from "Not a problem" to "A very serious problem. " The summary score ranges from 17 (no distress) to 102 (severe distress). The 25-item Interpersonal Processes of Care survey measures patients' perceived quality of their relationship with their providers; scores range from 1 (poor relationship) to 5 (good relationship). We also examined a single question regarding self-monitoring of blood glucose from the Summary of Diabetes Self-Care Activities questionnaire: "On how many of the last seven days did you test your blood sugar the number of times recommended by your health-care provider?"; responses ranged from 0 to 7.
Statistical Analyses
From the collected data, we created several additional variables of interest. The first was the summed injection frequency across insulin medications; if insulin injection frequency was missing, a once daily frequency was assumed. If method of insulin administration was missing, we assumed that a syringe was used. We classified patients as using a syringe, pen, or pump for insulin administration; syringe was reported if multiple methods were used (i.e., syringe and pen or syringe and pump). Only 12 patients were using an insulin pump alone; these patients' insulin administration type was excluded from the present analysis.
We performed descriptive analyses (mean ± standard deviation for continuous data; number and percentage for categorical data) for baseline variables of interest, for all patients and then by individual country. Due to the observational nature of the MOSAIc study, baseline data were not available for all patients for all variables of interest. The median proportion of missing data for any one variable was 4.9 %, interquartile range, 3.6–7.8 %. Patterns of missingness appeared to vary by country. Based on these facts, we used two approaches in our analyses. In the first, a "complete case" analysis, we calculated the descriptive measure among those patients for whom a value was present for a given variable of interest. Our second approach used multiple imputation via chained equations to accommodate continuous as well as categorical variables. Ten imputed datasets were created. The multiple imputation models included all variables of interest for which complete data were available as well as indicator variables for each country to account for country-specific variation. Imputation was done for the variable with the smallest proportion of missingness first and then in order of increasing missingness for other variables with missing data. To report descriptive statistics using the imputation approach, categorical and continuous values were converted to the log scale to approximate a normal distribution, averaged across the 10 imputed datasets, and then converted back to the linear scale; therefore, values and proportions for each level of the categorical values may not sum to the total number of patients or to 100 %. All analyses were performed using SAS 9.3 (SAS Institute; Cary, NC) and Stata 13 (StataCorp LP; College Station, TX).