Discussion
The management of rib fractures continues to be challenging. Clinical outcomes for this injury are multifactorial and present a real dilemma for prediction models that use independent factors. Management varies across a wide spectrum, and although we have found that many patients do not suffer complications, understanding the morbidity and mortality in those who do suffer complications is necessary for providing preemptive care. A successful model would allow physicians to identify patients most at risk of pulmonary complications and intervene early to reduce adverse outcomes.
A validated method to predict outcomes in this setting would be a powerful tool. The aim of our study was to compare three separate models and their ability to predict outcomes in elderly patients with rib fracture. Blunt chest trauma from falls is unique to elderly patients, and rib fractures are a common sequela of this mechanism. There are physiological, anatomic and pathological features that elderly patients share that differentiate them from younger cohorts. For example, we know that mortality from blunt chest trauma nearly doubles in elderly patients compared with younger ones. Moreover, with each additional rib fracture in an elderly patient, the risk of pneumonia increases by 27% and mortality increases by 19%.3 A validated scoring system, therefore, is of particular interest in this population and was the focus of our study.
The age that defined our elderly population (aged ≥55 years) may not appear to meet the social norm of what most people consider to be elder. Prior literature, however, suggests that even relatively young patients may be at a higher risk of complications from rib fractures. For example, Perdue et al demonstrated that age-related morbidity increased in non-elderly patients.11 Similarly, studies by Holcomb and Easter have demonstrated that patients over the age of 45 and even those as young as 40 exhibit increased mortality from rib fractures, respectively.12 13 Based on the review of a series of published data, guidelines at our institution recommend that patients aged ≥55 years with more than one rib fracture be admitted to an ICU monitored bed. For these reasons, we chose age 55 as the cut-off for our elderly population.
Using patient age as a factor to drive medical decision-making, as described above, is common across healthcare. Chronological age by itself, however, is not necessarily indicative of a patient’s overall state of health. There are many physiological factors that drive patient outcomes, and alternative measures that account for these may lead to more accurate prediction models. Frailty, for example, is a measure to quantify a patient’s physiological age as opposed to their true chronological age.
Our study compared three potential models to predict respiratory complications in patients with rib fracture. These models considered anatomic features of the rib fractures in addition to physiological factors such as frailty and arterial blood gas values. The anatomic model we used, known as the RibScore, has been previously proposed by Chapman et al to predict respiratory complications in patients with rib fracture.9 The mFI and arterial blood gas values are also known to be effective risk assessment tools to predict morbidity and mortality for patients in the surgical and trauma setting.4 10 14
Chapman’s study analyzed the CS individually for pneumonia, respiratory failure and tracheostomy in a younger cohort of patients compared with ours (median age 48 years vs 57 years, respectively). Patient age was shown to positively correlate with risk of complications, but incorporating age into the RibScore resulted in no improvement in its discriminative ability. Their CS was slightly higher in patients with isolated chest wall injuries (0.71 vs 0.77 for pneumonia, 0.72 vs 0.83 for respiratory failure and 0.75 vs 0.87 for tracheostomy). In contrast to Chapman’s study, we analyzed the composite ROC AUC CS for patient’s developing any one or more of these three complications. We found the CS for all patients in our population to be 0.79 for the RibScore. We think that a patient’s physiological age (or frailty), as opposed to their chronological age, is a more sensitive indicator of respiratory complications and that incorporating this into the RibScore would improve its discriminative ability.
Our results showed that mFI is an independent predictor of pulmonary complications in the elderly population and revealed a small, but not negligible, increase in its discriminative ability when compared with the RibScore (CS 0.83 vs 0.79, respectively). This suggests that a model based on frailty may perform at least as equally well as the RibScore. Furthermore, it shows that in addition to the anatomic features that compose the RibScore, frailty is a significant risk factor for predicting pulmonary complications in this patient population.
Initial PaCO2 showed the highest discriminative ability of the three individual models (CS 0.88). A combination of all three models yielded the highest CS of 0.90. This result suggests that initial PaCO2 may serve as a better individual predictor of adverse outcomes than models based on injury severity (ie, RibScore) and frailty. Arterial blood gases are performed routinely in cases of blunt chest trauma, and thus the data is easily obtained in the acute trauma setting.
It is recognized that each of the three models used in our study have their own individual limitations. With these limitations in mind, our goal was to combine models based off anatomic and physiologic variables to increase their predictive value. The RibScore may be limited by several factors, including radiologic interobserver variability. It may also be limited by the selection of patients who undergo a CT scan. For example, severely injured patients who are unstable (presumably with high RibScores) would be unlikely to receive a CT scan during their initial presentation. This may, in part, explain the low number of patients with a RibScore of 5 and 6 in both our present study and Chapman’s study.9 Although the mFI is useful in that its variables are standardized in the NSQIP database, it still relies heavily on thorough and accurate documentation. Of the three models, PaCO2 is the most objective, and this may account for it performing the best of the three individual models.
The three outcomes addressed in our study also have their own limitations. The decision to perform a tracheostomy was based on clinical judgement of the attending physician which may be a source of variability. Our data also does not stratify the timing of intubation, which could obscure pCO2 values for patients who may have been intubated prior to arrival. Additionally, unplanned intubation lasting less than 48 hours, although not defined as respiratory failure in our study, may still represent a significant pulmonary complication.
The scope of our study is also limited by a relatively small population. Although our findings are promising, our results need to be validated with a larger, multicenter study with the data set partitioned into both derivation and validation portions. Our current results can be considered hypothesis generating, and we are currently working on a prospective validation of our findings.
In conclusion, the RibScore maintains discriminative ability in elderly patients with blunt trauma rib fractures. However, models based on mFI and PaCO2 individually outperform the RibScore, and a combination of all three models yields the strongest discriminative ability. Our results suggest that a composite risk assessment tool based on RibScore, initial PaCO2 and mFI can be used to identify patients at risk of respiratory complications. This may serve as a preemptive tool for physicians to intervene early and reduce adverse pulmonary events in elderly patients with rib fracture.