Annual visits to GPs by elderly patients
CMAJ 1998;158:299
Re: "The health of Canada's elderly population: current status and future implications" (CMAJ 1997;157[8]:1025-32 [full text / résumé]), by Drs. Mark W. Rosenberg and Eric G. Moore
See response from: M.W. Rosenberg & E.G. Moore
See also:
Drs. Rosenberg and Moore set out to address the health status of Canada's elderly population and its impact on health care utilization. Unfortunately, they have added little to our current understanding and may in fact have created some confusion.
First, the authors could have used the NPHS in a more meaningful and creative way. Most of the tables and figures were adapted from published resources, so the paper does not present any new findings. The results are reported in an inconsistent manner because the authors were limited by the formats of the original publications. For example, the data in the tables and Figs. 1 and 2 were for people aged 55 years and over, presented in 10-year intervals, whereas those in Fig. 3 were for people aged 50 years and over, presented in 5-year intervals. No statistical tests or 95% confidence intervals were presented for any of the data.
Second, there are problems with the multiple logistic regression results shown in Table 5 (which appears to be the only table based on the authors' own analyses). Might the negative "odds ratios" in this table be regression coefficients that should have been further manipulated to generate the real odds ratios? It is disturbing that these values are then discussed as if they really were odds ratios.
The puzzling results are also reflected in the statistically significant associations between decreased GP consultations and some chronic conditions, such as arthritis or rheumatism and back problems. Apart from the problem with the odds ratios, the authors include disability in their model and treat it as a confounder. Disability is an intermediate variable lying between chronic conditions (cause) and increased GP consultations (effect). Including it in the model means that the true association between chronic conditions and increased health care utilization would be artificially underestimated. This is especially true for arthritis as the leading cause of disability.
Finally, the authors fail to describe how they selected variables for and fitted the logistic regression model. Some of the included variables have a very low frequency: for example, among those aged 65 and over, only 22 reported epilepsy. Not only is the inclusion of low-frequency conditions such as epilepsy in conflict with the NPHS statistical analysis guidelines, but it also contributes noise to the model. The authors do not state the age range to which the model applies: aged 50 and over, aged 55 and over, or aged 65 and over? Other questions include whether age was used as a categorical variable (nominal v. ordinal) or a continuous variable, how income was defined and which level was used as the baseline, and how weight and design effects were treated in the regression.
We tried to replicate the results shown in Table 5 of the published paper using the 1994-95 NPHS data set, but the agreement between our findings and those of Rosenberg and Moore was disappointing.
P. Peter Wang, MD
PhD student
Department of Public Health Sciences
University of Toronto
Toronto, Ont.
Elizabeth M. Badley, PhD
Professor
Department of Public Health Sciences
University of Toronto
Toronto, Ont.
 |
Send a letter to the editor responding to this letter
Envoyez une lettre à la rédaction au sujet de cette lettre |
Letters