Increased mortality following telemonitoring in frail elderly patients: look before you leap!

Abstract

T elemonitoring is often proposed as an efficient way to provide health care. The recent study by Takahashi et al examining telemonitoring in vulnerable patients with mixed chronic diseases clearly reflects the need for meticulous scientific approaches to study these types of interventions. Telemonitoring aims at early detection and prompt action in the case of health deterioration. Although patients reported high satisfaction and a sense of safety, telemonitoring failed to reduce hospital admissions and emergency department visits. Surprisingly, it resulted in a 4-fold increase in mortality risk (relative risk, 3.8; 95% CI, 1.3-11.0). This suggests that telemonitoring in frail elderly patients is hazardous, causing more harm than good. However, one can question the validity of this conclusion. A well-considered interpretation of the observed increased risk of mortality among patients receiving telemonitoring requires crucial information on timing and causes of death, which is currently lacking. The combined end point analysis ignores the true time-related impact of the exposure on mortality and health care utilization. In addition, it would have been informative to compare between-group characteristics of fatal cases vs nonfatal cases and indications for hospital admissions and emergency department visits. Despite randomization, it is not clear if both groups were comparable regarding their baseline mortality risk. An important constraint to obtain unbiased effect estimates in a randomized controlled trial (RCT) is that comparison groups are equivalent in terms of prognosis. It is well-established in statistical literature that hypothesis testing is inappropriate to evaluate differences in the distribution of baseline patient characteristics between treatment groups in RCTs. Nevertheless, the authors decided, based on P values, that both groups were balanced and adjustment of potential confounders was not necessary. It needs to be emphasized that even nonsignificant (P .05) imbalances of strong prognostic factors may still result in substantial bias and therefore requires adjustment. For example, chronic obstructive pulmonary disease, diabetes mellitus, and congestive heart failure were not statistically imbalanced between the treatment groups and yet are important risk factors of mortality and hence potentially confounding the effects of telemonitoring. These questions actually reflect the largest drawback of the study: the lack of substantial insight in the assumed relation between patient characteristics, intervention, and outcome. In intervention testing, the RCT is the final step, following a sequence of steps from initial preclinical research through phase 1 and phase 2 studies. The study by Takahashi et al warrants careful consideration of the benefits of telehealth interventions. Moreover, it shows the need of careful development and testing of nonpharmaceutical interventions.