**Biostatistics**** **

**Data Analysis Report **

** Introduction – Diabetes Control in the Elderly Project (DCEP)**

Diabetes affects at least 20% of the U.S. population over the age of 65. Evidence indicates that hyperglycemia (high blood sugar level) is associated with an increased risk of diabetic complications. The treatment of diabetes in the elderly is often hampered by age and disease related co-morbidities. Many physicians treat diabetes in the elderly without aggressive attempts to achieve optimal glycemic control. The primary goal of this project was to study whether improved glycemic control could be achieved in elderly patients without undue risk for hypoglycemia (severely low blood sugar).

Materials and Methods

128 elderly patients from several primary care clinics were randomized into either an intervention group where patients participated in intensive diabetes education, social support and aggressive attempts using agreed upon practice guidelines to improve glycemic control; or to a control group in which patients continued to receive their usual standard of care.

Eligible patients had to fit the following criteria to participate:

- a diagnosis of Type II diabetes
- age 65 or older
- no diagnosis of any organ failure, dementia or malignancy
- sign informed consent

Eligibility for participation did not depend on the number of years since diagnosis of Type II diabetes or on the presence/absence of diabetic complications (except those listed above). At the time of study entry as part of the consent process, patients agreed to the use of insulin if it became necessary and agreed to participate in all recommended services.

At the baseline visit, the patient’s age, sex, years since diagnosis, current treatment with insulin (yes/no), and their hemoglobin A1c (HbA1c) level were recorded. At the year 1 visit, current treatment with insulin (y/n), and the hemoglobin A1c were again recorded. In addition, the occurrence of any hypoglycemic incidents (1+/none) severe enough to require assistance, during the study year, was recorded.

Hemoglobin A1c (HbA1c), recorded as a percentage, is interpreted as a measure of average blood sugar level during the preceding 3 months and is a commonly used measure of glycemic control in diabetes. A low value is indicative of good glycemic control (lower blood sugar levels), and higher values of poor control (high blood sugar levels). A common goal for diabetic patients is to maintain an HbA1c level below 8%. Decrease of a half unit (0.5%) is considered clinically important in decreasing risk of complications.

Data is available on 113 patients who completed the study year. There was no difference in loss-to-follow-up rates in the intervention or control groups. A code sheet describing the data is included below.

Note: These data are a subset of a larger set of data used in a more complex and thorough analysis. Do not report on these data or use them in any other setting than completing this exam.

Research Questions

- Are there any differences in baseline characteristics of patients assigned to the control and intervention groups?
- Is baseline HbA1c level associated with number of years since diagnosis?
- Is baseline HbA1c level associated with baseline BMI?
- (a) Among the intervention patients, is there a significant ‘improvement’ in glycemic control over the study year? (b) Among the control patients?
- Is there a significant difference in change in glycemic control between control and intervention patients over the study year?
- Is occurrence of hypoglycemic episodes associated with treatment group?

Guidelines

Write a report for the investigators. Define and conduct hypothesis tests to address the research questions listed above. While there are other questions of interest to consider, keep your report focused on these questions. You should include any tables, summary statistics and figures that are useful for describing the data and presenting your results. Your hypotheses, assumptions, name of test or statistic used, results and conclusions should all be clearly stated. Be selective in presenting data and results. Raw (unedited) computer output is not acceptable (inclusion of output that has been edited for content and significant digits is encouraged). Your report is to be written for the investigators, and should include interpretation of the numbers and figures you present.

Data Code Sheet

Variable Description Code

DCID Study ID number sequential number

GROUP Randomly assigned 1=Intervention, 2= Control

Treatment Group

AGE Age in years ##.#

SEX Patient Sex 1=Male, 2=Female

YRSDIAG # of years since diagnosed ##.#

With Type 2 diabetes

PREA1c Baseline HbA1c (%) ##.#

YR1A1c Year 1 HbA1c (%) ##.#

PREBMI Baseline Body Mass Index ##.#

Index (BMI) in kg/m sq

PREINS Baseline Insulin Use 1=Yes, 0=No

YR1INS Year 1 Insulin Use 1=Yes, 0=No

HYPOGL Any episodes of severe 1=Yes, 0=No** **

**Solution**** **

**Data Analysis Report**

**Introduction – Diabetes Control in the Elderly Project (DCEP)**

Contents

Is baseline HbA1c level associated with number of years since diagnosis?. 5

Is baseline HbA1c level associated with baseline BMI?. 5

Is occurrence of hypoglycemic episodes associated with treatment group?. 10

### Are there any differences in baseline characteristics of patients assigned to the control and intervention groups?

No, at the moment of assigning to a certain group there were no baseline characteristics that were different. This is essential to be that way in order to be able to later make reliable comparisons between the two groups considering equal start.

### Is baseline HbA1c level associated with number of years since diagnosis?

The association between the number of years since diagnosis and the baseline level of HbA1c is tested with correlation analysis since both variables are continuous. The null hypothesis is that there is no association between the two variables.

According to the correlation analysis (Pearson Correlation) the baseline HbA1c is slightly positively correlated to the number of years since diagnosis, r = 0.252. Yet the correlation is significant at the 5% level (p-value = 0.007). The null hypothesis for no association is therefore rejected at the 5% level.

### Is baseline HbA1c level associated with baseline BMI?

The association between the baseline BMI and the baseline level of HbA1c is tested with correlation analysis since both variables are continuous. The null hypothesis is that there is no association between the two variables.According to the correlation analysis (Pearson Correlation) the baseline HbA1c is not significantly correlated to the baseline BMI, r = 0.136 (p-value = 0.150). The null hypothesis for no association therefore cannot be rejected at the 5% level.

Among the intervention patients, is there a significant ‘improvement’ in glycemic control over the study year? (b) Among the control patients?

The above research question is tested with a repeated measures ANOVA.

The assumptions of the repeated measures ANOVA are first checked:

- Dependent variable measured on a continuous scale: satisfied, the HbA1c (%) is measured on a continuous scale
- Independent variable should be categorical with at least 2 categories: satisfied, both time and group are categorical with two categories each
- No significant outliers: that is violated which can be seen on the table with large residuals below as well as on the histogram above

Fits and Diagnostics for Unusual Observations

Std

Obs A1C Fit ResidResid

42 12.000 9.440 2.560 2.31 R

66 12.500 9.423 3.077 2.78 R

72 12.200 9.423 2.777 2.51 R

81 11.800 9.423 2.377 2.15 R

126 11.200 8.584 2.616 2.36 R

137 11.900 8.584 3.316 2.99 R

155 11.500 8.584 2.916 2.63 R

168 10.900 8.584 2.316 2.09 R

179 12.000 9.330 2.670 2.41 R

213 12.100 9.330 2.770 2.50 R

213 12.100 9.330 2.770 2.50 R

- Relatively normal distribution of the dependent variable: that is tested with a histogram provided on the plot below and the assumptions seems rather satisfied

The goodness of fit of the model is very poor with an R-squared of 0.0936. That means that only 9.36% of the variability in the HbA1 could be explained by the time and group variables. The null hypotheses are that each of the true population coefficients is equal to 0.

The estimated coefficient for group 1 comparing the HbA1 from the pre and post period is positive (0.428) and significant at the 5% level (p-value < 0.0001). That means that on average the HbA1 level has increased with 0.428 from the pre to the post period for group 1. However, for group 2 this change is not significant (p-value = 0.661). Therefore in conclusion the glycemic control has significant improvement for the intervention group but not for the control group.

Analysis of Variance

Source DF AdjSSAdj MS F-Value P-Value

GROUP 1 7.506 7.506 6.01 0.015

time(GROUP) 2 21.131 10.566 8.46 0.000

Error 222 277.171 1.249

Total 225 305.809

Coefficients

Term CoefSECoef T-Value P-Value VIF

Constant 9.1945 0.0743 123.70 0.000

GROUP

1 -0.1823 0.0743 -2.45 0.015 1.00

time(GROUP)

1(1) 0.428 0.105 4.09 0.000 1.00

1(2) 0.046 0.106 0.44 0.661 1.00

### Is there a significant difference in change in glycemic control between control and intervention patients over the study year?

The difference between the two groups is significant at the 5% level (p-value = 0.015) and the estimated coefficient is -0.1823. That means that the level of HbA1 of the control group is significantly lower than that of the intervention group with 0.1823 on average.

### Is occurrence of hypoglycemic episodes associated with treatment group?

The association between treatment group and hypoglycemic episodes is tested with Chi-square and crosstab since those are two categorical variables.

The null hypothesis is that there is no association between the treatment group and the hypoglycemic episodes.

According to the table below the null hypothesis cannot be rejected at the 5% level (p-value = 0.247). Therefore the conclusion is there is no association between the two variables.

Rows: GROUP Columns: HYPOGL

0 1 Missing All

1 49 8 0 57

2 51 4 1 55

All 100 12 * 112

Cell Contents: Count

Pearson Chi-Square = 1.338, DF = 1, P-Value = 0.247

Likelihood Ratio Chi-Square = 1.363, DF = 1, P-Value = 0.243