Clinical and medication profiles stratified by household income in patients referred for diabetes care

Data Sources

To conduct this work, we used a regional DEC database that captures basic demographic information on all attendees to the regional clinic situated in Calgary, Alberta, a large Canadian city. The sampling frame was all active patients at the DEC from May 1, 2000 to January 9, 2002. The sample consisted of 4687 patients. All patients included were from a single health region within the province of Alberta. This DEC is the single regional provider of diabetes education services. Access is dependent upon physician referral to the centre. The postal codes of patients registered in the DEC database were linked to their corresponding dissemination area (DA) using the Statistics Canada Postal Code Conversion File (PCCF).

Neighborhood income data were obtained from Statistics Canada Census data (2001). We defined a neighborhood as equivalent to a census dissemination area (DA)- a small geographic area covered by a single census data collector which typically contains 400–700 persons. Therefore, median household income per DA was the income measure used in this study. These data were merged with the DEC database on the variable DA. Neighbourhood income has been shown to be reasonably concordant with individual income in urban settings [29, 30]. There is also increasing evidence that neighbourhood income is valid SES construct that predicts health outcomes independently of individual income [31, 32].

Derivation of Income Quintiles

Study Variables and Statistical Analyses

Physicians referring patients to the DEC complete a standardized referral form that includes clinical data. This information was then entered into the DEC patient registry. Clinical information examined in this study included: serum hemoglobin A1C (HBA1C); serum lipid profiles including levels of low density lipoprotein (LDL-c), high density lipoprotein (HDL-c) and triglyceride; microalbumin to creatinine ratios and medications used at time of referral. Height and weight are measured upon presentation to clinic; these measures were used to calculate the body mass index (BMI) which was then entered into the DEC database.

Potential differences in continuous clinical parameters across income quintiles were examined using regression models. If inspection of the distribution of these variables suggested a linear relationship between income and the variable of interest, then income quintile was modeled as a single ordinally-coded predictor variable. If, on the other hand, the relationship was not linear, then regression was performed using dummy variables for each income quintile relative to the lowest income quintile as a reference group. Covariates considered in these models included sex and medication use. Differences in categorically-coded medication use across income quintiles, meanwhile, were examined using Chi square analyses. All statistical analyses were performed in STATA, version 8.

Descriptive Analysis

Clinical characteristics of patients referred for diabetes care and education are listed, by income quintile, in Table 1. The median age of patients increased as income level decreased. The median age in the highest income groups (quintile 5) was 55.3 years compared to the lowest income group (quintile 1) which was almost 2 years older at time of referral with a median age of 57.0 years. There was a similar inverse relationship between income level and BMI. The median BMI in the wealthiest quintile (quintile 5) was 28 compared median BMIs of 29.6 in quintile 1 and 29.8 in quintile 2. Our results also indicate that the lowest income quintile presents to clinic later from the time of diagnosis of diabetes. The median duration of diabetes was 4 years in the lowest quintile, compared to 3 years in all of the other quintiles. In terms of diabetes-related clinical parameters, patients did not differ significantly across groups with respect to serum LDL-c levels. Patients in the highest income quintile had the highest HDL-c. An inverse relationship between income and triglycerides is suggested as the median triglyceride levels range from 2.40 mmol/L in the lowest income group down to 2.12 mmol/L in the highest income group. Glycemic control appears to be slightly better in the highest income group as evidenced by a median HBA1C of 8.4% in quintile 5 compared to a median HBA1C of 8.9% in all other income groups. There is also a suggestion of a negative association between microalbumin creatinine ratio (M:C) and income. The lowest income group had a median M:C of 2.3 compared to a median M:C of 1.5 in the highest income group. Boxplots illustrating the distribution of clinical characteristics are illustrated in figure 1.

	Income Quintile
	1 (low)	2	3	4	5 (high)	p -for trend*
Clinical Characteristic	Median (IQR)
Age (in years)	56.95 (22.9)	56.52 (21.48)	56.95 (19.94)	55.24 (19.44)	55.27 (18.5)	0.032
BMI	29.6 (8.6)	29.8 (8.3)	29 (7.9)	29.5 (8.2)	28 (7.2)	<0.0005
Duration of Diabetes (in years)	4 (9)	3 (8)	3 (10)	3 (9)	3 (7)	0.98
LDL-c (mmol/L)	3.01 (1.26)	2.97 (1.1)	3.02 (1.22)	2.97 (1.23)	2.99 (1.26)	**
HDL-c (mmol/L)	1.12 (0.41)	1.1 (0.36)	1.09 (0.37)	1.1 (0.36)	1.15 (0.35)	**
Triglycerides (mmol/L)	2.40 (1.88)	2.29 (1.99)	2.41 (2.0)	2.32 (1.76)	2.12 (1.61)	**
HBA1C (%)	8.9 (3.7)	8.9 (3.7)	8.9 (3.5)	8.9 (3.7)	8.4 (3.3)	**
Microalbumin: Creatinine	2.3 (9)	1.95 (7.7)	2.4 (7.1)	1.6 (5.9)	1.5 (5)	**

*results of univariate analyses for trend when income could be modeled as a single, ordinally-coded variable

**these variables did not have a simple linear relationship with income (results of regression models where income quintiles are modeled as dummy variables are displayed in Table 3).

Regression Analysis

Clinical and Biologic Factors

This study demonstrates that there are clinically significant differences in some biologic parameters across income quintiles and that low income patients present to clinic with higher risk profiles. Low-income patients are older at time of referral and have more atherogenic metabolic profiles with higher serum triglycerides and lower HDL levels, which are associated with a higher risk for developing cardiovascular disease [33–35]. While a significant association between income and HBA1C was not noted, inspection of the distribution of HBA1C suggests a trend towards a lower HBA1C in the highest income quintile.

This study also suggests that differences in metabolic status are not due to overt under-treatment of the economically disadvantaged. The lowest income groups were using more sulfonylureas and metformin compared to the wealthiest groups. Even the use of more costly therapies such as TZD and lipid-lowering therapies were similar across income groups.

Behavioural Factors

This study also provides some insight into potential health related behavioural differences across income groups. It has been shown in previous research that sedentary lifestyles are more common among lower income populations. In this study, the lower income groups had the highest BMIs. This raises the possibility that the lower income groups are less physically active than their wealthy counterparts. The lower HDL levels and higher triglycerides might also reflect behavioural differences with respect to diet and/or exercise [36]. Unfortunately, the data set used in this study does not contain detailed data on diet or exercise so we were unable to explicitly assess differences in these important behaviours.

Other considerations

High income is frequently associated with higher health literacy and a greater ability to apply health-related knowledge [37, 38]. It should be noted that while we did not find that HBA1Cs differ significantly at the time of referral, others have documented that individuals from higher socio-economic strata are more likely to experience significant lowering of their HBA1Cs after assessment at diabetes clinic [37, 38]. It would be most interesting to know had the clinical profiles of these patients been re-evaluated 1 year following their referral whether the differences noted would have remained the same, been attenuated or perhaps been even more pronounced due to differences in health literacy.

While we did not find a significant difference with respect to the duration of the diagnosis of diabetes at the time of referral, examination of the distribution of this variable certainly suggests that this may, in part, be mediating some of clinical differences noted. The wealthiest patient group was also younger, and more likely to be controlled with diet alone, suggesting that these patients may be presenting at an earlier point in the natural history of their diabetes. If wealthy patients were being referred earlier (perhaps due to earlier diagnosis), this may also help explain the inverse relationship between income and complication risk. As there is now clear evidence that aggressive management of blood glucose, high blood pressure and high serum lipids will effectively prevent the micro- and macrovascular complications of diabetes [39–43], it follows that the earlier a specialist intervenes, the more effective these prevention strategies might be.

This study has limitations. This is a cross sectional study that examined the clinical profiles of patients at one point in time. These referrals were not necessarily index referral, and had we compared clinical profiles at first contact with specialty care, it is possible that some of the clinical differences noted may have been attenuated. It is noteworthy that clinical data were entered into the DEC database from a standardized clinic referral form. All clinical data examined in this study, therefore, were provided by the referring physician. If doctors differ in the manner in which they complete, or do not complete this form, an information bias could be introduced to this study. We do not have any evidence, however, that physicians' documentation skills should differ based on the neighbourhood income of their patients, and would assert that information bias relating to income is unlikely.