Prognosis for Mechanically Ventilated Patients: A Moving Target.

Invasive mechanical ventilation remains fundamental to the management of critically ill patients in the ICU. As the worldwide population ages and develops an increasing number of medical comorbidities, rates of mechanical ventilation are also rising (1, 2).

Within this context, prognostic information regarding mechanically ventilated patients is increasingly important for patients and their surrogate decision-makers. In the ICU, shared decision-making, or medical decisions made through a partnership among physicians, patients, and their loved ones, is the recommended standard (3). However, patients, their surrogates, and physicians often have different expectations regarding prognosis, with prior data showing a >50% discordance between surrogates and physicians (4). A key pillar of shared decision-making is being able to provide patients and their surrogates with reliable expectations. Previous studies have attempted to predict mechanical ventilation outcomes at specific time points (Day 1, Day 14, and Day 21) with the assumption of static prognoses (5–7). However, patients receiving mechanical ventilation change from day to day and static prognoses at predetermined time points may not be appropriate for an ever-changing population.

In this issue of AnnalsATS, Ruan and colleagues (pp. 729–735) used data from 162,200 episodes of respiratory failure included in Taiwan’s National Health insurance database to investigate dynamic changes in mechanical ventilation prognoses based on each additional day of mechanical ventilation needed (8). The authors identified adults who received mechanical ventilation for two consecutive days, and calculated the cumulative probabilities of weaning success and death in the subsequent 90 days. Their results showed that >90% of successful weaning occurred in the initial 30 days after mechanical ventilation, with a decreasing trend over time. In contrast, deaths initially increased after mechanical ventilation, but then decreased after the 19th day on the ventilator, with the probability of death surpassing the probability of weaning success on the 28th ventilator day. The authors’ findings were consistent across multiple subgroups.

Based on their results, the authors created an online inquiry system to provide tailored prognostic information based on ventilator day, age, and sex (http://mvp.nhri.org.tw/NHIA-NHRI2017/count.html). They believe that this information may provide patients and surrogates with more dynamic information regarding evolving prognoses that may impact decision-making in the ICU.

Although Ruan and colleagues used a large database to provide potentially helpful prognostic information for critically ill patients and surrogate decision-makers, their study has some limitations. First, there is a concern about the generalizability of the results, as this study was performed in a single country, Taiwan. It is unclear whether the etiology of respiratory failure, ventilator practices, and treatment protocols that may affect outcomes correlate with those in other regions. Notably, palliative withdrawal of life support did not occur in Taiwan during the time of this study. Because overall mortality is high in ICUs, palliative withdrawal of life support occurs in many countries for various reasons, including patient/family preference, imminent death regardless of intervention, conditions incompatible with long-term survival, and conditions leading to an unacceptable quality of life (9). If palliative withdrawal of life support had been included in this study, it likely would have altered the cumulative probability of weaning success and the probability of death. Palliative withdrawal may occur for a number of reasons, but often occurs in individuals with a low weaning potential. If patients who are less likely to have successful weaning die early owing to palliative withdrawal, the patients left alive would likely be those with a higher chance of weaning success, which would increase the probability of successful weaning. Similarly, in a population where some patients die after transitioning to a palliative approach, the probability of early death would likely increase and the probability of death might decrease during later phases of mechanical ventilation. Therefore, although the authors’ findings may hold true for the Taiwanese population, the specific probabilities of death and weaning success would likely differ in countries with different practice patterns. This difference could have a significant impact on decision-making outside of Taiwan.

In addition, although the authors stratified their analyses across multiple subgroups, primarily based on comorbidities and the presenting diagnosis, they did not account for severity of illness. Severity of illness is a strong predictor of death, perhaps even stronger than the clinical factors included in Ruan and colleagues’ study (10), and is a key consideration in ensuring that patients and their families receive the most robust prognoses about death and weaning success. Sicker patients are more likely to die and less likely to be weaned. However, severity of illness is a dynamic marker: a patient may initially be very sick, but after treatment, certain markers of severity may improve. As such, Ruan and colleagues’ approach involving dynamic prognoses would be synergistic with the inclusion of dynamic measures of illness severity.

Finally, the premise of this study rests on the concept that patients and surrogates will find the dynamic probability of weaning success and death to be informative in the process of shared medical decision-making. Investigators should use this tool to determine whether that is truly the case, or whether patients and surrogates will choose to remain optimistic about outcomes despite being presented with these data.