Performance Status in Cancer: Not Broken, But Time for an Upgrade?

BACKGROUND

Considerable advances in early detection and combination therapy during the past 70 years¹ have contributed to significant cancer-specific survival gains, with an estimated 17 million individuals currently living with a history of cancer in the United States, a number expected to reach 26 million by the year 2040.² The financial investment to achieve such progress, of course, has been enormous.^3,4 Intriguingly, despite the increasing sophistication and cost of contemporary cancer care,^5,6 evaluation of patient performance status (PS)—an integral aspect of treatment selection, toxicity monitoring, and clinical trial eligibility—has remained essentially unchanged since 1948.⁷ In this commentary, we provide a historical overview and critical evaluation of PS assessment in oncology. We also discuss alternative approaches to PS assessment that may improve prognostication and risk stratification in research and clinical practice.

PS SCALES: 72 YEARS AND COUNTING

In 1948, Karnofsky and colleagues⁷ used three criteria to evaluate nitrogen mustard efficacy in lung cancer: objective improvement (eg, decrease in lesion size), subjective improvement (eg, patient-reported symptoms), and PS (ie, patients’ ability to participate in activities of daily life). Evaluation of PS was standardized using a physician-rated scale, now known as Karnofsky Performance Status (KPS), ranging from 0 (dead) to 100 (well functioning) with 10-point increments.⁷ The simpler 6-point Eastern Cooperative Oncology Group (ECOG) PS scale (0 [fully active] to 5 [dead]) was introduced by Zubrod and colleagues⁸ in 1960 as one of 15 standardized assessments for all ECOG multicenter clinical trials. The intended purpose of the ECOG PS scale was to evaluate patient reaction to chemotherapy, along with patient-reported pain, nausea, and appetite.⁸ Both clinician-administered scales were used primarily to evaluate therapeutic efficacy until 1973, when Zelen concluded that failure to consider PS in clinical trial eligibility “will introduce so much variability and bias into the trial that real differences between therapies are likely to be missed altogether.”^9(p34) As a result, subsequent use of PS scales in clinical trials transitioned from the assessment of therapeutic efficacy to eligibility (ie, KPS ≤ 60 or ECOG ≥ 2 ineligible)^10-13 and stratification (ie, more homogenous subgroups).^14-16 Both the KPS and ECOG scales are strong independent predictors of clinical outcome in numerous oncology populations.^17-19 They are also inexpensive and feasible to implement in all oncology clinical settings. Consequently, the KPS and ECOG scales, and therefore the assessment of PS, remain integral tools in contemporary practice incorporated into virtually every clinical visit across the entire cancer care continuum.

KPS AND ECOG: ROOM FOR IMPROVEMENT

The demonstrated clinical value of KPS and ECOG directly contradicts why one would advocate for their replacement. Nevertheless, closer inspection of the KPS and ECOG scales reveal significant limitations. First, these scales are clinician based and therefore subjective, with poor reliability and validity.²⁰ In a meta-analysis of 15 studies representing 2,808 patients, agreement between clinicians on KPS/ECOG scores was deemed moderate (Pearson correlation coefficients, 0.71 to 0.78).²¹ In addition, patient and clinician PS scores have low agreement,²² with up to 50% of patient-reported functional limitations missed by clinicians²³ and adverse PS change (ie, KPS < 60) reported approximately 15 months earlier by patients than clinicians.²⁴ Misclassification of PS has obvious implications for clinical trial eligibility as well as planned best practice therapy, with some patients classified as having sufficient PS to tolerate therapy but having nascent impairment, and conversely, another proportion classified as having insufficient PS yet with considerable reserve capacity and able to tolerate therapy.^25-27 Second, poor KPS and ECOG scores are strong predictors of prognosis,^17-19 which is not surprising when impairment is obvious; however, the prognostic value in patients with good PS is limited. A meta-analysis of 66 phase II and III randomized controlled trials (n = 44,511 patients with ECOG 0-2)¹⁸ found no differences in clinical outcomes between patients with either ECOG scores of 0, 1, or 2, suggesting that ECOG provides essentially no additional prognostic information for patients with no obvious physical impairments. Third, KPS/ECOG may have limited use for toxicity risk stratification in the era of contemporary practice. Use of PS scales for toxicity risk prediction is based on work in the 1980s that showed a higher incidence of chemotherapy toxicity in patients classified with poor PS (ie, KPS ≤ 60; ECOG ≥ 2).^11,26,28 The safety profile of current treatment regimens, however, has evolved considerably, in part because of the use of supportive polytherapy care, the safety profile of modern scheduling approaches/lower-dose combinations, and the use of molecularly targeted agents.²⁹ Indeed, a retrospective study of 16,233 patients with solid tumors receiving contemporary chemotherapy regimens found negligible differences in relative dose intensity between patients with ECOG PS of 0 versus 1-3.³⁰ Finally, KPS/ECOG scales are only administered during in-person clinic visits, often weeks, if not months, apart; therefore, a snapshot of PS is captured, which limits the ability to detect more subtle real-time changes of potential clinical importance.

Overall, given the highlighted limitations, it is somewhat remarkable that PS scales exhibit any clinical value. We posit that the reason KPS/ECOG have widespread applicability is because they provide insight into PS, a metric with substantial importance in clinical populations. PS measurement, albeit using a variety of different assessment tools, is of central importance in virtually every area of clinical disease management.³¹ We further suggest that more objective, discriminatory, and dynamic tools may actually augment the value of PS assessment in the oncology setting beyond that currently possible with KPS and ECOG scales.

ALTERNATIVE PS MEASURES

Alternative tools with which to assess PS or aspects of PS have been explored in the oncology setting (Table 1). For instance, the geriatric assessment³² (patients age > 65 years) and the hematopoietic cell transplantation–specific comorbidity index³³ (hematologic malignancies) are multiparametric tools that evaluate different PS domains (eg, activities-of-daily-living questionnaire) and/or comorbidity indices (eg, cardiac disease). Preliminary data suggest that both tools improve toxicity and survival prediction beyond PS,^32,33 leading to widespread clinical uptake, at least in certain oncology settings. Both tools are limited, however, as they are population specific and typically assessed at one time point.³⁴ Therefore, pan-cancer assessment tools that provide an objective and dynamic PS evaluation must be identified.

TABLE 1.

Exemplar Performance Status Assessment Tools

Numerous standardized tests that include counting repetitions or timing an activity have been developed to evaluate PS in noncancer clinical populations. The Sit-to-Stand Test,³⁵ Timed-Up-and-Go test (TUG)³⁶ and Short Physical Performance Battery³⁷ were developed to assess lower-extremity strength, balance, and mobility in frail elderly individuals. In patients with cancer age ≥ 70 years, both poor TUG test³⁸ and Short Physical Performance Battery³⁹ have been associated with an approximately two-fold increased risk of mortality. The 6-minute walk test, originally developed for chronic obstructive pulmonary disease,^31,40 is a robust predictor of mortality in respiratory and cardiac diseases,^31,41 whereas preliminary data in patients with cancer indicate a 6-minute walk test result of < 350 meters predicts mortality risk, even after adjustment for KPS.^42,43 However, these tools fail to discriminate between patients who are classified with good PS⁴⁴ and will likely not detect or identify therapy-related toxicity.^41,45

Cardiorespiratory fitness (CRF), first quantified by Hill and Lupton in 1923,⁴⁶ provides an objective assessment of the integrative capacity of the cardiovascular, pulmonary, hematopoietic, and musculoskeletal systems to transport and use oxygen.³¹ The gold standard assessment of CRF is a cardiopulmonary exercise test coupled with automated gas exchange assessment, which provides direct quantification of submaximal and peak oxygen consumption (VO₂peak).³¹ Prediction equations that are based on achieved exercise workload can also estimate VO₂peak. In chronic lung and cardiac conditions, CRF is considered a clinical vital sign, with assessment recommended across all phases of clinical decision making.^31,47,48 In cancer, Reichel⁴⁹ first reported in 1972 that presurgical estimated CRF was a stronger predictor of postoperative complications compared with standard clinical metrics (eg, age and pulmonary function) in lung cancer, concluding that CRF “should be used routinely in the evaluation of the candidate for pneumonectomy.”^49(p576) Subsequently, the tolerability (ie, more than 80% of patients achieve peak criteria) and safety (ie, no exercise-related deaths and an approximate 15% nonserious adverse event rate) of CRF was established in a systematic review of 90 studies representing approximately 5,200 patients.⁴⁵ Work in lung,^50,51 GI,⁵² hepatobiliary and pancreatic,⁵³ and hematologic malignancies⁵⁴ show that CRF is an independent predictor of postoperative complications and mortality.^55-59 Presurgical CRF stratification values are defined for lung⁴⁷ and colorectal cancer,⁵⁸ with less than 15 mL O₂⋅kg⁻¹⋅min⁻¹ associated with an elevated risk of complications, whereas less than 10 mL O₂⋅kg⁻¹⋅min⁻¹ is considered to be associated with a high risk of complications.⁴⁷ Poor CRF is also associated with a higher prevalence of acute and chronic treatment-related late effects^55-59 and all-cause, cardiovascular, and cancer mortality.⁶⁰ Despite major advantages, CRF requires specialized equipment and trained personnel, likely limiting its widespread clinical application.^48,61

In summary, despite promising findings, none of the aforementioned tools is currently incorporated into routine clinical practice. In addition to the dearth of evidence that supports the clinical use of such tools, a fundamental weakness of all tools is the stark discrepancy in practicality compared with the KPS/ECOG scales. There is a critical need for tools that can dynamically and objectively assess PS with the widespread feasibility of the KPS/ECOG scales.

THE DIGITAL FRONTIER

Advances in health technology may provide unprecedented opportunities to improve PS assessment in cancer. Most previous and ongoing studies leveraging technology focus on monitoring objective mobility via wearable devices. For instance, accelerometers are small, wireless, digital devices providing objective mobility measurement (eg, steps per day and/or minutes per day of light, moderate, and vigorous activity).^62,63 Saint-Maurice et al⁶⁴ found that, compared with less than 4,000 steps per day, more than 8,000 steps per day was associated with a lower risk of all-cause and cancer-specific mortality in 4,840 apparently healthy individuals. Preliminary findings from two studies in patients with advanced disease⁶⁵ and those undergoing surgical tumor resection⁶⁶ indicate that lower steps per day and total activity per day are associated with increased risk of death and postsurgical complications, respectively. Few studies are performing deep, dynamic phenotyping using multiparametric platforms; however, the rapid expansion of consumer-grade digital mobile devices—for example, Fitbit, Apple iWatch, smartphones—with integrated multisensory systems may offer a unique opportunity for digital phenotyping. Such devices dynamically generate an abundance of unlabeled sensor data points per day, including mobility and physiologic data, such as heart rate and sleep.^67-70 The feasibility of digital phenotyping was recently demonstrated in work showing that 250,000 daily measurements in 43 individuals over 11 months were used to first develop personalized, activity-based baseline normative data, and abnormal phys-iologic signals from longitudinal data were subsequently used to identify early signs of disease.⁷¹ An ongoing prospective cohort study in 2,500 patients with heart failure (ClinicalTrials.gov identifier: NCT03810638) will create a digital registry that combines patient-reported outcomes (PROs), mobility, and electronic health record data to monitor changes in quality of life and other clinical outcomes. These are examples of the type of digital phenotyping studies needed in oncology. We posit that the integration of digital technologies assessing physiologic (eg, heart rate), mobility (eg, steps per day), and PRO (eg, PRO-Common Terminology Criteria for Adverse Events),⁷² as well as electronic health record data, may be of significant value for treatment selection, clinical trial eligibility, and toxicity monitoring in the oncology setting.

Several challenges remain to realize the potential of digital phenotyping in the oncology setting. Policies governing the privacy and security of patient data captured from digital mobile devices will need to be defined. Research efforts will require development and validation data sets and robust statistical tools, such as discrimination, calibration, and reclassification, to discern whether digital phenotyping–related data provide incremental value beyond the KPS/ECOG scales. Subsequent randomized trials that assess whether baseline/change in digital biomarkers improve current patient stratification approaches for informing clinical decision making are needed. Data repositories that allow for the simultaneous integration, query, and visualization of large amounts of heath data with artificial intelligence tools^73,74 are needed to seamlessly integrate digital phenotyping data into clinical workflows.⁷⁵ Overcoming such challenges is a US Food and Drug Administration priority, with efforts underway to advance the standardization of PS data in oncology.⁷⁶

In conclusion, the prognostic utility and widespread applicability of the KPS/ECOG scales reflect the central importance of PS in patients with cancer. Nevertheless, the complex and heterogeneous nature of cancer management has accentuated the need for the development and validation of objective and dynamic measures of PS that can accurately discriminate between patients across the continuum of cancer care in all settings. The clinical impact of such measures would be considerable.

AUTHOR CONTRIBUTIONS

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Performance Status in Cancer: Not Broken, But Time for an Upgrade?

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).

Elisabeth Edvardsen

Employment: GlaxoSmithKline

Lee W. Jones

Stock and Other Ownership Interests: Pacylex

No other potential conflicts of interest were reported.