CORR Insights®: Can a Psychological Profile Predict Successful Return to Full Duty After a Musculoskeletal Injury?

Where Are We Now?

Counseling patients regarding expectations before and after treatment is a critical component of shared decision-making. As Greenlee et al. [6] found in this month’s Clinical Orthopaedics and Related Research®, this is true for injured soldiers and athletes of all types when considering the risk of reinjury after return to normal activity. This topic has been studied before, but the relationship between reinjury and patient psychologic profiles remains debated. This study approaches this question with a modern and much-needed perspective in orthopaedic surgery. In general, our practices are moving toward a “precision medicine” approach, understanding that the “median isn’t the message” [4]. The authors [6] present a modern statistical approach to prediction modeling for reinjury after return to active duty to guide patient-specific counseling after musculoskeletal injury. The authors make an important distinction between “mental health” and the many components of mental health, such as depression, anxiety, mood, and pain catastrophizing. Surgeons must consider the multifactorial nature of mental health when treating patients with musculoskeletal injury and disease, and surgeons should be cognizant of psychologic attributes beyond depression and anxiety.

Where Do We Need To Go?

Prediction of future, recurrent musculoskeletal injuries with only psychologic profiles is difficult, as shown by the authors in this study [6]. They note that future studies must collect other, nonpsychologic factors including details of the injury itself if one is to build accurate prediction models. It is crucial that we continue to investigate mental health domains as we move forward and hope to create individualized care pathways. This is particularly true for potentially modifiable variables, like depression, anxiety, and pain catastrophization. Patient-reported outcome studies in musculoskeletal disease often fail to consider that mental health is a composite outcome of many different psychologic domains and health states. The authors [6] made the important distinction between a number of these domains, though some measurements were made with factors that have not been shown to be valid in any situation (such as, “rate your level of personal satisfaction with job/military/life”). We need to revisit the concept of mental health in patients with musculoskeletal disease and approach analyses with a more detailed understanding of where deficits lie and how they affect outcomes, both subjective (for example, pain) and objective (such as reinjury).

The authors [6] used simple instruments to collect data on some mental health domains, using the validated Patient Health Questionnaire-9, Total Pain Catastrophizing Scale, and Fear-Avoidance Beliefs Questionnaire. From an academic perspective, I believe it is important that the groundwork be laid with the use of validated health-related quality of life instruments. Broad adoption, and thus clinical utility, will come with instruments that are easy to administer, thereby reducing question burden, and are easy to score, encouraging use in all practices. We must identify which factors are important. Is it depression? Is it anxiety? From there, we will need shorter instruments that still offer an accurate estimate of disease burden (such as the severity of depression). The Patient-Reported Outcomes Measurement Information System offers hope for this, with computer adaptive tests and quick scores that can be interpreted relative to the population mean [3].

Many situations in musculoskeletal disease are multidimensional, including the risk of reinjury. It is likely that age, index injury and recovery, expected level of activity, pain perception, psychologic attributes, and other unknowns affect the probability of reinjury. The study in this month’s CORR® [6], with a prediction model generated using Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis reporting guidelines, underscores the complexity of these analyses and is a first step in the correct direction. Machine learning is sometimes offered as a panacea for all our statistical needs and shortcomings. Although this is not the case, machine-learning ensembles excel when data are multidimensional and offer us the opportunity for improved prediction models, offering probabilities at the individual patient level. It is not generally useful to cite odds ratios to patients in clinic because these neglect other factors that are important to an accurate prediction. We must continue to investigate and “fail” with model generation if we are to make progress in the future.

How Do We Get There?

This study [6] is an early step in what will be a long road toward precision medicine for patients with musculoskeletal disease. We need more large observational cohort studies, collecting subjective data (for example, validated patient-reported outcome instruments) and objective data (such as demographic and surgical) using continuous variables, when available. As we become more facile with research drawn from electronic medical records, we will have larger amounts of data at our fingertips available for analysis (and in the future, for personalized predictions of outcomes). Large amounts of multidimensional data may allow machine-learning techniques to shine, while they do not seem to offer much for smaller datasets [9]. Our desire is for categories (such as age < 65 or ≥ 65 years or BMI measured as obese, morbidly obese, and so on), and we often fail to recognize that data granularity is lost with such blunt categorization of continuous variables [1, 5]. For example, there’s likely to be little difference between a patient with a BMI of 34 kg/m² and someone with a BMI of 35 kg/m² in terms of surgical risk, though those two patients might fall on opposite sides of some arbitrary cutoff. Patient-reported outcome measurements should be validated to ensure we are accurately measuring the condition (such as depression) and ensure that patient-reported outcome measures are responsive to change. Observational cohort studies will inform us about which psychologic traits are most detrimental to musculoskeletal health outcomes and worthy of further investigation.

Once we understand the relationships between psychologic health domains and outcomes in musculoskeletal care, we can explore interventions (for example, cognitive behavioral or pharmaceutical therapies) to try to modify those psychologic health domains. The current assumption is that these are modifiable patient-level attributes and that modification will improve outcomes, thereby providing greater value. Without a fundamental understanding of the relationships between psychologic health and outcomes and with categorization of continuous variables (for example, depressed versus not depressed), we will continue to miss the mark by performing research that will not move us forward, at the expense of time, money, and our patients. If we perpetuate prior mistakes and do not learn from previous work, we will fail to achieve the grail of personalized medicine [7].

Read This Next

• A column by Stephen J. Gould [4] is a nice look into the differences of population-level research versus the precision-medicine approach we must take moving forward. Although averages (and odds ratios and relative risks) are useful from a “mile-high” perspective, they are not particularly useful when counseling any individual.

• Another opinion piece [2] offers insight into the complexity of researching “mental health,” which is multidimensional and challenging to investigate. Focused work in a dimension will likely help us make more progress than continued, broad-based work that fails to acknowledge complexity.

• A book [8] offers concrete examples that underscore the difficulties we face with decision-making in general. Informed decision-making is made even more complex in the setting of large amounts of data because of the noise that exists in our biological systems. Perfect decision-making algorithms may never exist, although we will likely offer more useful information to patients with better analyses and an understanding of the influence of noise on an particular outcome of interest.