Long-term survival of elderly patients after intensive care unit admission for acute respiratory infection: a population-based, propensity score-matched cohort study

The ageing of the population is a global trend of critical importance, with the number of individuals aged 80 years or older increasing the most rapidly (3.8% increase/year) [1]. This worldwide ageing has major consequences on health systems including increasing elderly patient admissions in hospitals and intensive care units (ICUs) [2, 3]. ICU admission of elderly patients is controversial, but randomised controlled studies are difficult to design due to ethical limitations [4]. Indeed, conflicting results are present in the literature: some studies report that advanced age is itself a risk factor for ICU mortality and poor long-term quality of life due to persistent functional impairments [5‐8], while other studies argue that chronological age alone should not be considered the sole criterion to preclude ICU admission [9, 10].

Most studies focusing on post-ICU effects among the elderly have considered elderly patients as a single entity. However, we feel it is questionable to discuss hospital admission policies for patients based solely on their advanced age without considering the acute disease leading to admission. Discussions of these topics should be disease-specific, and a focus on respiratory infections is of particular interest because of the strong association of these diseases with old age [11]. Consequently, we previously demonstrated a substantial increase in hospitalisations for acute respiratory infection (ARI) over 10 years (2006–2015), with a change in the ICU admission policy for elderly patients leading to an important rise in ICU resource utilisation [11]. For example, the number of nonagenarians hospitalised in ICU for ARI was 5.8-fold higher in 2015 than in 10 years earlier. However, the long-term effects of ICU admission for elderly patients with ARI have never been demonstrated.

Assessing the burden of post-ICU effects among elderly patients is particularly difficult due to the impossibility of comparing ICU-treated individuals with the most appropriate control group. Comparing elderly patients hospitalised for ARI according to whether or not they were admitted to the ICU is importantly biassed: non-ICU patients include patients that did not meet the severity criteria for ICU admission along with other cases that are sometimes severe but received only palliative care. Comparing ICU-treated patients from different age groups also does not provide useful information because of the increased risks of death and functional impairment that clearly accompany increased age.

The objective of this study was to examine the 2-year outcomes among elderly patients discharged from the hospital after being admitted to the ICU for ARI. To describe the gap in morbi-mortality between the ICU survivors and the theoretical medical condition expected at this age, we studied differences between elderly survivors of ARI and a control population representative of the physiologic effects of ageing, and matched on age, sex and preexisting comorbidities. The primary outcome was mortality, and the secondary outcomes were healthcare utilisation and frailty score over the 2 years after discharge.

We demonstrated that patients over 80 years old hospitalised in ICU for ARI were discharged alive from the hospital in 75% of cases. However, these patients had a 10-fold increased risk of death 6 months post-hospitalisation and 3.6-fold increased risk of death at 2 years post-hospitalisation as compared to a control population. Further, during the 2-year post-ICU period, healthcare utilisation and frailty score doubled among the ARI patient group.

We performed a propensity-matched population analysis of patients with ARI admitted in ICU compared to patients undergoing cataract surgery, which was assumed to have no effect on survival or future re-hospitalisation. The insignificant mortality observed during the stays for cataract surgery combined with the comparable pre- and post-healthcare utilisation and frailty score evaluations confirmed this population as a relevant control. Importantly, matching was performed based on comorbidities that were present before the critical illness of ARI patients. As the populations were comparable in terms of age, sex and underlying diseases before the acute illness, one can attribute the excess mortality observed after hospital discharge to the severe illness and associated treatment (ARI requiring ICU hospitalisation) and not to spontaneous ageing or consequences of preexisting underlying conditions. Our study shows that an elderly patient being discharged alive from ICU is clearly not the end of the story, but rather an important turning point in life. The main findings of this study could be expressed from binary viewpoints. On one hand, as a group, a substantial proportion of elderly patients will die in the 6 months after discharge from the ICU. Moreover, the survivors will endure a critical decrease in their quality of life, as suggested by the worsening of the frailty score and the increase in healthcare use we observed after hospitalisation. On the other hand, on an individual level, many elderly patients surviving ICU hospitalisation will live more than 4 years after discharge (239 patients among 988). These findings highlight the need for early consideration of the management of frailty and newly developed comorbidities after ICU discharge, and more generally for management of the overall post-ICU rehabilitation process. Taking the management of hip fractures in elderly patients as an example, close cooperation between emergency physicians, anaesthesiologists, intensivists, surgeons, geriatricians and rehabilitation physicians has proven to drastically reduce the risk of death in the months after hospital discharge [19, 20]. ARI is a severe and frequent disease among the elderly and requires the same attention. Another finding of our study is that 6 months is a more appropriate time frame to define the true ICU mortality for an elderly population, rather than the usual 30-day mortality [19, 21].

The present study has several strengths. First, we defined an inventive control population and matched based on preexisting conditions present before the occurrence of critical illness. By using medico-administrative databases, we were able to compare pre- and post-hospitalisation trajectories and healthcare utilisation. Previous cohort studies have provided substantial knowledge on the long-term outcomes after hospital discharge [22‐24]; however, prospective cohort studies are limited in the amount of information they can provide from long-term follow-up because of the lack of prehospital trajectory data [25]. Moreover, many of these long-term follow-up studies lack appropriate control groups for comparison. Consequently, there are uncertainties regarding the extent to which the morbidity and mortality observed after hospital discharge in these studies can be attributed to the critical illness and the extent to which preexisting underlying conditions contributed to such findings (which is a major limitation for elderly populations with expected comorbidities and a high mortality rate). Second, our study used a large contemporary database of 39 acute healthcare facilities in which collection of data on every hospital stay is mandatory. These data are generated based on the patient’s routine care without any intervention affecting the usual methods of patient care. This study design could be assimilated as a real-life study, which contributes to the high generalisability of the results [26]. Third, in addition to survival analyses, this study assessed healthcare utilisation and frailty score evolution, which are important and relevant outcomes in an elderly population. The frailty score measures susceptibility due to the age-associated decline in reserve and function in a wide range of physiological systems [27]. Frailty was associated with an increased risk of mortality in critically ill patients older than 80 years [28] and generally in elderly patients [29]. The doubling of the frailty score that we observed in the ICU survivors is consistent with the doubling of healthcare utilisation observed over the same time frame.

This study also has several limitations. First, the use of administrative hospital databases introduced an inherent bias that should be taken into consideration. The strengths and limitations of using healthcare databases for epidemiological purposes have already been extensively discussed [30‐34]. Second, the mortality was estimated based on readmissions and mortality at hospital. Out-of-hospital deaths were not recorded, and we were thus unable to calculate the actual mortality. Based on data from the “Fin de vie en France” [End of life in France] survey, we know that dying people in France do not stay in their own homes and mostly end up in a hospital: only 14% remains at home throughout the last month of life [35]. Additionally, we expect that the rate of out-hospital deaths was similar in our experimental and control groups. Indeed, our algorithm captured the vast majority of the deaths, and we can assume that the observed increased risk of death is accurate. Third, we cannot completely exclude the possibility that there were factors affecting only one group (ARI or cataract group) that were not measured in the data available for matching. The use of two approaches (matching and pre/post analysis) demonstrated consistency and was against this latest hypothesis. However, it should be noted that only the main comorbidities, as opposed to all comorbidities, were used in matching (Table 2). The matching process also may have excluded patients with more severe illness (see Table 1). This may have led to an underestimation of the gap between morbi-mortality of elderly survivors of ARI in the ICU and theoretical values expected for this age group (represented by the control cohort). Moreover, patients were not matched on socioeconomic status and residential status (assisted living, etc.), both of which would have at least some impact on the dependency measures. Fourth, it is likely that triage based on the appropriateness of ICU admission for elderly patients was very selective. Elderly individuals are admitted to ICU only if a high likelihood of survival is expected a priori. Consequently, our results can only be extrapolated to health systems with equivalent elderly triage processes. In particular, this study is specific to the French healthcare setting and may not be generalisable. Finally, we did not include elderly patients with ARI who were admitted in the general ward as controls. Having an intermediate group of patients who survived ARI without ICU admission would have conferred additional strength to this study and given insight to any potential “dose-response” where higher illness severity potentially increases the long-term burden on elderly patients recovered from acute illness. However, patients were highly heterogeneous in the general ward and were comprised of those (i) not severe enough for ICU admission and (ii) meeting the criteria for ICU admission but who were too old and/or had too much comorbidity. The survivors in this heterogeneous population are difficult to characterise, because their reasons for ICU non-admission cannot be captured by the ICD-10 algorithms. Moreover, these two populations (“elderly in general ward” and “elderly in ICU”) were by definition different, because the physicians in charge of each patient initially made a call as to whether ICU admission was appropriate. Thus, trying to match these two groups by a propensity score is likely inappropriate. A prospective randomised controlled trial is probably the ideal method to answer the question of ICU benefits and the burden of post-ICU effects, but ethical concerns limit the feasibility of this approach.

Our data provide insight into outcomes during a 2-year follow-up of patients over 80 years old who were discharged alive from ICU after severe respiratory infection. We demonstrated an important increase in the rate of death in the years after hospital discharge along with elevated healthcare resource use and accelerated age-associated decline as assessed by the frailty score. These findings provide data for more informed goals-of-care discussions and may help target post-ICU discharge services for these high-risk patients.