Article Text


Effect of dichotomising age in multivariate model analysis
  1. A M Yohannes1,
  2. M J Connolly2
  1. 1Manchester Metropolitan University, Elizabeth Gaskel Hatersage Road, Manchester M13 0JA, UK; a.yohannes{at}
  2. 2Department of Geriatric Medicine, University of Auckland, Auckland, New Zealand
  1. J J Soler-Cataluña3,
  2. M A Martínez-Garcia3,
  3. P Román Sánchez3
  1. 3Hospital General de Requena, Unidad de Neumología, Servicio de Medicina Interna, Requena, Valencia 46430, Spain; jjsoler{at}

Statistics from

We read with interest the paper by Soler-Cataluña and colleagues1 that examined—in an impressive prospective study with 5 years follow up—factors predicting poor prognosis and mortality in patients with severe acute exacerbations of chronic obstructive pulmonary disease (AECOPD). Their findings are complementary with the current available literature in identifying that older age, arterial carbon dioxide tension, and acute exacerbations were independent predictors of mortality in their cohort group.

We have concerns, however, regarding both their analyses and conclusions. Firstly, several studies2–4 have given advice on the limitations of dichotomising continuous predictors as they come at a cost “as explanatory variables could be seriously misleading, both in respect of which variables are significant in the model, and perhaps also with respect to the overall predictive ability”.2 Soler-Cataluña and colleagues state that in their multivariate model “the frequency of acute exacerbations, age and Charlson index were analysed as categorical variables”.1

Secondly, and perhaps more importantly, the authors have reported older age (clearly a non-modifiable factor) as a predictor of death. They do not state whether they believe this to be old age per se or an age related potentially modifiable variable. Have the authors collected data on social support, physical disability, depression, quality of life, and any palliative care their patients may have received during the follow up period? These variables may have some effect on mortality in this exclusively male COPD patient cohort. Our own group has recently published data on 1 year mortality following hospitalisation for AECOPD in a slightly older group of subjects (mean age 73 years v 71 years in the patients studied by Soler-Cataluña and colleagues) with worse baseline spirometry (mean percentage predicted FEV1 39%). In our study age was not a mortality predictor on either univariate or multivariate analysis. Quality of life, level of disability, severity of depression, readmission, use of long term oxygen therapy, and duration of original admission (all of which are arguably related to age) were all univariate predictors of 12 month mortality, with only the quality of life score remaining a significant predictor on multivariate analysis.

We wonder whether the inclusion of age related variables in the study by Soler-Cataluña et al, together with the use of age as a continuous variable, might have resulted in qualitatively or quantitatively different conclusions regarding the effect of age on prognosis. However, the inclusion of duration of original admission and of frequency of readmission in our own list of predictors5 would support their suggestion that severe AECOPD could have an adverse impact on longer term mortality.


Authors’ reply

We wish to thank Dr Yohannes for his interest and comments on our study1 and have the following comments on the questions he raises.

Firstly, although it is true that in some cases the transformation of continuous variables into dichotomised variables my induce some changes in the results obtained, in other cases the use of continuous data may conceal some partial effect, particularly if the predictive relation is non-linear. In fact, in our study the only age group to show a poorer prognosis were those aged ⩾75 years (odds ratio (OR) 5.26, 95% CI 2.70 to 10.24). In the same way, categorisation of the number of exacerbations allowed us to review the differential effect of repeated exacerbations. For these reasons, and in order to make interpretation of the results easier, we decided to apply categorisation of some continuous variables in our study. On the other hand, we should mention that this transformation of variables did not modify the results, as both age and the number of exacerbations behaved as independent prognostic factors on inclusion in the model as continuous variables. Specifically, in this predictive equation, age proved to be an independent prognostic variable with an OR of 1.06 (95% CI 1.01 to 1.11). The same applies to the number of exacerbations with an OR of 1.20 (95% CI 1.03 to 1.39).

Secondly, with regard to the role of age as a predictor of mortality, different studies involving both stable patients2 and acute cases3,4 have also found age to be an adverse prognostic factor. Despite such evidence, we consider the hypothesis suggested by Yohannes—that other age related and potentially modifiable variables would determine the prognostic effect attributed to age—to be very interesting. Unfortunately, in our analysis we did not include measures such as social support, physical disability, depression, or quality of life so we are unable to assess their specific weight. Almagro et al,5 in a study that also explored mortality predictors after hospitalisation and which considered variables of this kind, found age to have a predictive value in the univariate analysis, but this effect disappeared in the multivariate study. Therefore, as suggested by Yohannes, it is probable that the effect of age may be minimised when other predictors that condition or define such an effect are included in the model.

In conclusion, age dichotomisation did not substantially change the results and conclusions drawn in our study. Re-analysis of the data using continuous (non-dichotomised) variables continues to suggest that severe exacerbations are independent predictors of mortality.


View Abstract


  • Competing interests: none declared.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.