Statistics used to track patient diets. (Data Management & Analysis).
Evaluating competitive products, the researchers chose SYSTAT from Systat, Inc., Richmond, Calif., because of its core and add-on capabilities.
The study consisted of an initial descriptive phase, a second phase of improving problems identified in the initialphase, and a third phase in which improvements in clinical practice would be seen. The third phase also included a randomized controlled trial with a control group and an intervention group. Altogether, data was gathered on about 2,000 patients.
Initial data collection
Data on 750 patients was collected randomly among all newly admitted patients. The purpose of this phase was to examine whether departments receiving patients after admission would screen them for nutritional problems, and if problems were detected, how they were dealt with.
The method of identifying nutritional problems was a screening method that included information about body weight, recent weight loss, recent dietary intake, and the severity of disease. This screening resulted in identifying patients as being nutritionally at-risk or not at-risk. To determine how patients were categorized, a logistic regression analysis was carried out in SYSTAT.
This analysis was followed by a model prediction success table showing that 95% of the patients were correctly classified by a logistic regression equation, revealing that not only were all components necessary, but also that the result of the analysis had a close relationship to observed data.
In the descriptive part of the study, an analysis of length of stay was needed to evaluate whether a longer length of stay among patients with nutritional problems was related to the nutritional problems identified, or rather to other variables recorded.
This analysis was performed with a generalized linear model analysis in SYSTAT. A number of continuous and categorical variables were entered as independent variables, and then an analysis carried out by an automatic complete analysis or by a manual/automatic stepwise regression. The results revealed that the length of stay was related to nutritional problems.
Finalizing the results
After the initial phase, a period was spent on issuing guidelines, teaching and improving hospital food. After this a follow-up recording of another set of 750 patients was made. Key indicators showed significant improvement. Clinical outcome was evaluated by the length of stay suitable for this study.
Since several factors are known to influence length of stay, it was decided to analyze the results in subgroups. After completion of the study, the data was analyzed by two-way ANOVA (analysis of variance), which revealed an interaction between group (control or team) and complications.
In addition, length of stay in the team group of patients with complications appeared to be shorter than the length of stay in the control group of patients with complications. Here some help from the SYSTAT developers was needed, since post-testing is not an option directly after a two-way ANOVA.
The developers recommended a generalized linear model analysis with the same categorical variables, but including the interaction term as an independent variable, followed by a pairwise comparison.
Since the generalized linear model analysis allows testing of specific hypotheses, testing whether the whole group of control patients differed significantly from the whole group of team patients was also done. In two-way classified data with unequal numbers of observations in cells (the unbalanced or non-orthogonal case), the correct least squares means which estimate the linear model row or column parameters, are not the naive row or column (weighted) means, but simple averages of the cell means. Thus to an untrained person it might appear that what a statistical package outputs as adjusted least squares differences are wrong, since it may not bear any similarity to the observed differences, say between control and team.
However, comparisons of row or column effects are best made with these least squares estimates of means. If the numbers of observations in cells are all the same, both these types of estimates will give the same values. SYSTAT computes proper adjusted least squares means, whether the cell sizes are equal or unequal.
Initially Systat's results appeared mysterious to us, but with the above explanation from the SYSTAT developers, this mystery was resolved; there was no significant difference in means between the groups.
Systat Software, Inc., 510-231-4786, www.systat.com
--Jens Kondrup, nutrition unit, Rigshospitalet, Copenhagen, Denmark
|Printer friendly Cite/link Email Feedback|
|Publication:||R & D|
|Date:||Jun 1, 2003|
|Previous Article:||DATs rally. (Data Management & Analysis).|
|Next Article:||Deforming meshes adapt to modeling motions. (Data Management & Analysis).|