Healthcare trajectories before and after critical illness: population-based insight on diverse patients clusters

In this population-based study, we comprehensively tracked healthcare use of ICU survivors admitted for ARDS and/or septic shock ventilated at least 5 days, 2 years before and 2 years after ICU. Using a clustering strategy based on pre-ICU healthcare use, we could depict the different healthcare trajectories nested in the global population of ICU survivors. Global pre-ICU healthcare trajectory revealed a progressive increase in healthcare use, but clustering analyses showed that most of the patients had a stable low healthcare use until ICU admission or no healthcare recourse.

Interestingly, Szakmany et al. [17], in a recent population-based data linkage study, similarly reported a high healthcare facility use the year before ICU admission, and Lone et al. [10] also observed in their population-based study a global increase in healthcare use before ICU admission. Hence, one may consider that acute critical illness is the consequence of a global worsening medical condition, and consequently, post-ICU increased healthcare use and PICS might be its consequences rather than causally related to the acute critical illness and ICU stay per se. The present work identifies various trajectories hidden in this global picture. The methodology of the present work, using the exact same methods to assess healthcare use of the pre- and post-ICU periods, enabled a paired analysis, each patient being its own control. Thus, patients with low and stable or without any pre-ICU healthcare use, which represented the vast majority of patients (83%), had a persistent and stable increased healthcare use over the 2 years post-ICU. This suggests that the acute critical illness episode had a key role in the healthcare use trajectory of most survivors who have moderate and stable healthcare use before, resulting in a higher level of “basal” healthcare use after the ICU stay. These results could be regarded as measurable consequences of PICS at the population level. Moreover, major comorbidities related to renal, respiratory and cardiac functions were more frequently reported in the post-ICU period compared to the pre-ICU period, also supporting this view. Regarding the group of patients with elevated healthcare use in pre-ICU period, our clustering analysis identified two groups with different dynamic of healthcare use utilization. Thus, group B displayed a steep increase before ICU admission, probably as a result of worsening underlying medical conditions, eventually leading to acute critical illness. Group D, however, displayed a decrease before ICU admission, which appears unexpected following a period of high healthcare use. We cannot rule out the possibility—although unlikely—that these patients were referred to other health structures outside of the study region, or referred only to private ambulatory physicians unrelated to private hospitals, both being uncaptured by our extraction algorithm. We can also hypothesize that it represented patients lost for follow-up after an important period of healthcare use, and consequently readmitted later in critical condition. Underpinning this hypothesis, this group had a large proportion of patients with chronic respiratory and cardiac diseases, medical conditions for which close follow-up is mandatory to avoid unscheduled emergency readmissions [22, 23].

These results are of interest in the current new era of critical care providers’ concern about long-term outcome of their patients: a better profiling and understanding of healthcare use trajectories of critically ill patients will help understand patients most likely to benefit of ICU admission in the long term and give valuable information for setting up follow-up programs. Interestingly, differences in healthcare use observed in pre-ICU among the five groups remained consistent in post-ICU period. Moreover, despite these important differences in level of healthcare use, there were no obvious differences in main characteristics of the ICU stay between groups (Table 3). In the same line, Lone et al. [10] showed that prior illness and healthcare use were stronger predictor of post-ICU hospital readmissions than the acute critical illness by itself. Thus, such pre-ICU healthcare trajectories analyses may be of interest in future works to improve long-term outcome of critically ill patients. Integrating pre-ICU healthcare trajectories to select patients for post-ICU follow-up programs could, therefore, be helpful and more discriminant than characteristics of the ICU stay itself. Moreover, we might hypothesize that dedicated post-ICU follow-up interventions will differentially affect patients, depending on their specific healthcare use trajectory prior to ICU admission. Specifically, targeting patients with a previous stable and low healthcare for post-ICU follow-up might be more efficient, with recovery objectives easier to define.

At a public health level, for the evaluation of global ICU-induced costs, this work paves the ground for future detailed analyses and evaluations according to healthcare use trajectories. Previous landmark studies have demonstrated elevated costs associated with ICU admissions, compared to non-ICU control populations [10, 24]; however, differences in case-mix preclude detailed analysis and interpretation. Our work shows that relevant sub-groups among ICU patients with prior moderate and stable healthcare use could be defined, and be relevant to settle global pre-ICU, ICU and post-ICU healthcare costs, enabling comprehensive cost/effectiveness assessment of ICU admission and potential post-ICU follow-up programs.

Last, at a conceptual level, the observed switch in healthcare use trajectories after critical illness, with a persistent increased healthcare use after critical illness can also be analysed using the theoretical frameworks of the critical transitions and loss of resilience, that has recently gained considerable interest in many science fields, including systems biology [25, 26]. One could, thus, interpret critical illness as the tipping point, with further frailty and low resilience state during the post-ICU period, with measurable consequences in terms of frequent healthcare use.

This study has several limitations. First, we used administrative databases which could be at risk for miscoding or undercoding. The French national hospital database used here was initially designed for billing purposes but now appears to be a powerful tool for epidemiological surveillance on the condition that the selection algorithm was validated [27‐30]. We assessed the performance of our selection algorithm through blinded review of medical charts randomly selected and we observed a very good performance of the detection (positive predictive value 96%, negative predictive value of 92%). Second, during the post-ICU period, patients who were lost to follow-up and who died might bias data interpretation. However, we aimed at limiting this bias (1) by restricting our analyses to patients living in the region during the study period, and, thus, unlikely to have healthcare use outside the region—not captured in the study and (2) by using quarterly measurements of healthcare use reported as a ratio on the total days spend alive with follow-up for each patient considered analysed. Third, data extraction was limited to hospitals (public and private); thus, healthcare use associated with individual practitioners not affiliated with a hospital was not evaluated in our study, leading to a possible under-estimation of healthcare use and/or lack of capture of associated specific healthcare use trajectories. However, one may assume that the most severe cases of healthcare recourse lead to hospitalization at one point, or, at least, an ambulatory care/specialist consultation affiliated with a hospital, all captured in the present study. Last, we could not extract data regarding frailty, which has recently emerged as an important and relevant concept in critical care [31, 32].