Treatment Outcomes in Patients with Migraine: An Ex-Post-Facto Comparison of Two In-Patient Facilities

Abstract

Background: In two hospitals we performed an open, prospective observational study on patients with chronic headache as a measure of internal quality assurance using identical methods. Available data were subordinately analysed in order to compare both studies. Questions: Are the patient samples of both hospitals comparable? If not, which form of statistical adjustment is recommended? Are there differences in the outcome measures of both facilities? Methods: The outcomes were defined as differences between baseline values and values at discharge from hospital, respectively 6 months after. Frequency of headache attacks, intensity of pain, intensity of general complaints as well as of concomitant symptoms, and quality of life were determined in advance as dependent variables. To compare both patient groups univariate analysis of variance without and with inclusion of covariates were used. For calculating propensity scores (conditional probability of belonging to one of two groups) a logistic regression with the same covariates serving as independent variables was performed. Patients: 426 patients with the main diagnosis ‘Migraine’ and complete data sets concerning the covariates were selected for analysis. 87% of patients are female, the mean age is 45.5 ± 11.7 years (range 14–73 yrs). Results: 4 out of 11 potential covariates show statistically significant differences between the patients of both hospitals. Correct classification of patients by means of the propensity score succeeds in 67%. Comparing the outcomes at discharge from hospital, significant differences between both groups exist which are, with one exception, not affected by controlling for covariates. 6 months after discharge two of the initial differences between both patient groups are no longer present. These findings are independent from the statistical technique of risk adjustment. Conclusions: Because of the observed differences between both patient groups it is recommended to adjust data by regression analysis in order to enhance comparability. The choice for one of the two proposed techniques is secondary. With respect to the analyses clear differences between both hospitals exist in short-term outcomes, disappearing 6 months later.

Profiling quality of care for patients with chronic headache in three different German hospitals – a case study

“…By performing analyses of covariance we chose a linear regression analysis based model in order to adjust for these differences. We did not calculate propensity scores [ 15 ] because both methods resulted in similar conclusions [ 16 , 17 ]. Instead, we considered the definition of the relevant adjustment factors and the choice of the indicators of outcome quality to be more important.…”