- Email: [email protected]
Mortality by a proxy performance status as defined by a claims-based measure for disability status in older patients with newly diagnosed multiple myeloma in the United States
JGO-00663; No. of pages: 7; 4C: Journal of Geriatric Oncology xxx (2019) xxx
Contents lists available at ScienceDirect
Journal of Geriatric Oncology
Mortality by a proxy performance status as defined by a claims-based measure for disability status in older patients with newly diagnosed multiple myeloma in the United States Shuling Li a,⁎, Tanya Natwick a,1, Jiannong Liu a, Vicki A. Morrison a,b, Sarah Vidito c, Winifred Werther d, Akeem A. Yusuf c, Saad Z. Usmani e a
Chronic Disease Research Group, Hennepin Healthcare Research Institute, 701 Park Avenue, Suite S4.100, Minneapolis, MN 55415, USA Hennepin Healthcare Systems, 701 Park Avenue, Minneapolis, MN 55415, USA c Amgen Inc, 1 Amgen Center Dr, Thousand Oaks, California 91320, USA d Amgen Inc, 1120 Veterans Blvd, South San Francisco, California 94-80, USA e Carolinas HealthCare System, 1021 Morehead Medical Drive, Charlotte, North Carolina 28204, USA b
a r t i c l e
i n f o
Article history: Received 2 August 2018 Received in revised form 20 December 2018 Accepted 8 January 2019 Available online xxxx Keywords: Administrative claims Cancer Medicare Mortality Population-based cohort Proxy for performance status
a b s t r a c t Objectives: We applied a claims-based definition of disability status as a proxy for performance status (PS) and examined associations between PS and mortality in a population-based cohort of older US adults with multiple myeloma (MM). Materials and Methods: We identified older (≥66 years) Medicare beneficiaries diagnosed with MM January 1, 2008-December 31, 2011, who began first-line therapy in the study period (through December 31, 2012). We estimated predicted probability of poor PS for each patient at initiation of each line up to fourth-line therapy, classified as poor (predicted probability ≥0.11) or good (b0.11) PS, and examined mortality. Crude overall survival was estimated using the Kaplan-Meier method with log-rank test for survival comparison between PS groups. Cox proportional hazards models evaluated the association between poor PS and mortality risk, adjusted for baseline characteristics by lines of therapy. Results: Of 12,547 patients, 5841, 2372, and 819 initiated second-, third-, and fourth-line in the study period. Poor PS proportions were 16.6%, 21.8%, 18.4%, and 18.2% at each line. Crude overall survival was worse for poor PS patients across lines (P b 0.01 for each). Adjusted hazards ratios (95% CI) of mortality for patients with poor versus good PS were 1.28 (1.18–1.40), first-line; 1.55 (1.36–1.77), second-line; 1.35 (1.10–1.65), third-line; 1.22 (0.84– 1.76), fourth-line. Conclusion: The claims-based prediction model for disability status performed as expected as a proxy for PS in older Medicare patients with MM. PS was an independent risk factor for mortality. Further studies assessing the effect of PS on mortality by therapies are warranted. © 2019 Published by Elsevier Ltd.
1. Introduction Multiple myeloma (MM) is a malignancy of predominantly older patients, accounting for 1.8% of all incident cancer cases and 2.1% of all cancer deaths in the US population [1]. In 2018, an estimated 30,770 new cases were diagnosed, and 12,770 patients died of MM [1]. MM
⁎ Corresponding author at: Chronic Disease Research Group, Hennepin Healthcare Research Institute, 701 Park Avenue, Suite S2.100, Minneapolis, MN 55404, USA. E-mail addresses: [email protected] (S. Li), [email protected] (J. Liu), [email protected] (W. Werther), [email protected] (A.A. Yusuf), [email protected] (S.Z. Usmani). 1 Currently affiliated with OptumLabs, Minneapolis, Minnesota, USA.
incidence increases with age, with N60% of diagnoses and 78% of deaths occurring in patients aged 65 years or older [1,2]. Over the past decade, several novel MM therapies have been approved, leading to substantial improvement in survival [3]. Older patients, however, benefit less from new treatments, possibly due to decreased treatment tolerance, increased health burden, and worse biology [4]. Patients with poor health status are especially at risk of treatment-related adverse events and poor clinical outcomes due to reduced physiologic reserve and limited adaptability to stressful events [5–7]. Performance status ([PS], e.g., Eastern Cooperative Oncology Group [ECOG] or Karnofsky Performance Status), one dimension of patients' health status, is defined by ability to carry out physical activities, some of which are related to self-care and independent living [8]. PS is used
https://doi.org/10.1016/j.jgo.2019.01.007 1879-4068/© 2019 Published by Elsevier Ltd.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
2
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
by clinicians to evaluate patients' health condition and assess risks and benefits in treatment choices. This information is usually not available in administrative claims data. Thus, observational studies using medical claims data to evaluate treatment patterns and compare the effectiveness of therapies on survival are often subject to bias due to unmeasured confounding, such as PS [9]. Although two claims-based PS algorithms have been developed using data from a single tumor registry for lung cancer [10] and from outpatient community oncology practices including ten tumor types [11], a claims-based definition of PS for older general Medicare patients with MM has not been established. Davidoff and colleagues developed an algorithm linking self-reported functional status measures based on activities of daily living (ADL) and instrumental ADL to a disability status scale (0–4, representing a progression of increasing degrees of limitation) using the Medicare Current Beneficiary Survey data linked with Medicare claims. This disability status scale was a proxy for the ECOG PS scale, and was dichotomized to focus on good (ECOG 0–2) versus poor (ECOG 3–4) PS. Davidoff et al. then identified potential claims-based predictors of poor PS and developed a prediction model for proxy PS, defined as poor or good based on a dichotomous measure for the predicted probability of poor PS in the general Medicare population [8;12]. The purpose of this study was to apply the claims-based measure of disability status developed by Davidoff et al. to a population-based cohort of older US adults with MM as a proxy for PS, and to assess the association between PS and mortality. 2. Materials and Methods 2.1. Data Source The Centers for Medicare & Medicaid Services database for the 100% sample of Medicare beneficiaries with hematologic cancers including MM from 2007 to 2012 was used in this study. The files include demographics, Part D low-income subsidy (LIS) status, Medicare enrollment status, dates of service, and diagnostic and procedure codes. Mortality information (date of all-cause death) was obtained using the date of death field from the Master Beneficiary Summary File. We applied to and received approval from the Human Subjects Research Committee of Hennepin County Medical Center/Hennepin Healthcare System, Inc. 2.2. Study design and Population This was a retrospective cohort study covering January 1, 2007, through December 31, 2012. The study population included Medicare beneficiaries diagnosed with MM between January 1, 2008, and December 31, 2011, based on a validated algorithm using a combination of International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes for MM (203.00, 203.01, and 203.02) and diagnostic tests or treatment [13]. The algorithm required patients to have an MM diagnosis (index diagnosis) and (i) three additional MM diagnoses during the 90 days before the index diagnosis AND either one bone marrow or two other diagnostic tests during the 90 days before the index diagnosis (Appendix A1) or (ii) chemotherapy within 180 days after the index diagnosis (Appendix A2-A4). The index diagnosis date was identified as the disease index date. We further required patients to be aged at least 66 years at the disease index date, to have initiated a first-line therapy following the disease index date (the date of treatment initiation defined as the first-line index date), and to be continuously enrolled in fee-for-service Medicare Parts A, B, and D from 12 months before the disease index date to the first-line index date. In an effort to include only patients with newly diagnosed MM, we excluded patients who had received any chemotherapy, radiotherapy, or stem cell transplant (Appendix B, D), or drug treatments specific to MM (Appendix C) in the 12 months before the disease index date. Patients with missing census region based on the state of residence were also excluded.
For each treated patient, first-line to fourth-line therapies were identified based on an algorithm derived from healthcare administrative claims and electronic medical record databases. The first-line therapy started on the first date of any MM-related chemotherapy. Appendix C provides a list of codes for drugs available during the study period. The first line ended at the earliest of the following: 1) a gap of N90 days in all MM-related therapies included for the first-line therapy; 2) addition of or switch to a new MM therapy (not including prednisone) that was not adjunctive to the first-line regimens; 3) death; or 4) disenrollment from any Medicare Part A, B, or D coverage or December 31, 2012. Second-, third-, and fourth-line therapies were defined using the same approach. Drug regimens were identified based on the anti-myeloma medications used within 90 days of the line of therapy index date using National Drug Codes from Medicare Part D prescription drug event claims and Healthcare Common Procedure Coding System codes from Part B line items and Part A outpatient claims (Appendix C) and classified as monotherapy, doublets, or triplets based on the National Comprehensive Cancer Network MM treatment guidelines [14]. Regimens not identified as one of these three types were classified as “other.” Prednisone alone was not considered monotherapy for MM, but was allowed to be part of a drug regimen. The baseline period for each line of therapy was the 12 months before the treatment index date of the line, during which PS, comorbid conditions, Deyo-adapted Charlson Comorbidity Index (CCI) [15], length of hospital stay (LOS), and LIS were defined. The follow-up period for each line of therapy was the corresponding line duration, and the overall follow-up period was from the first-line start date to the fourth-line end date. 2.3. Measures A claims-based poor disability status prediction model developed and validated by Davidoff et al. [8;12] was used to estimate the probability of poor disability status as a proxy measure for PS at line-oftherapy initiation for each patient via the following steps: First, we defined the healthcare service predictors as described by Davidoff et al. [8] from Medicare claims during the baseline period for each line-of-therapy cohort. Healthcare service predictors included evaluation and management visits, other visit types, minor or ambulatory procedures, major procedures, preventive services, durable medical equipment use, imaging, and other service indicators. The prediction model also included Medicaid enrollment status and sex, identified from the Master Beneficiary Summary file. Next, we applied the regression coefficients provided in Appendix C from Davidoff et al. [8] to the set of constructed measures for each patient to generate a predicted probability of poor disability status, with high values representing high probability. Finally, based on the predicted probability, patients were classified as poor PS (probability of poor disability status ≥0.11) or good PS (b0.11). The 0.11 cut-off was chosen by Davidoff et al. to maximize the best balance of sensitivity and specificity among a cohort of Medicare beneficiaries [8]. We performed a spline curve analysis of effect of predicted probability on patient survival and found no evidence to reject the use of this cut-off to identify patients with poor disability status used as proxy for poor PS in a cohort of older patients with MM. Demographics (date of birth, sex, race, geographic region) were identified in the Medicare Beneficiary Summary file with age calculated at each line-of-therapy index date and categorized into four groups (66–69, 70–74, 75–79, ≥80 years). Medicare Part D LIS was identified in the Medicare beneficiary enrollment file. Hospital LOS during the baseline period was defined from inpatient claims and categorized into three groups (0, 1–10, ≥11 days). Level of comorbidity was assessed during the baseline period using the Quan version of the CCI [15], and categorized into four groups (0, 1–2, 3–4, ≥5). Comorbid conditions not included in the CCI were also identified during the baseline period. Most comorbid conditions were defined using standard methodology
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
[16], and skeletal-related events were identified using the algorithm of “base case” definition from Aly et al. [17] All-cause mortality was assessed during the relevant follow-up period for each line of therapy cohort and overall throughout first-line to fourth-line therapy.
3
poor PS and time-varying confounders (regimens and comorbid conditions) updated at the subsequent line of therapy index date, since these factors may potentially predict PS and mortality and serve as intermediate variables on the causal pathway between poor PS and death.
3. Results
2.4. Statistical Analyses Baseline characteristics are described by PS and lines of therapy. We report counts and percentages for categorical variables, mean (standard deviation [SD]) and median (interquartile range [IQR]) for continuous variables. Mortality rates are expressed as numbers of patients who died per 100 patient-years. Crude all-cause survival was estimated using the Kaplan-Meier method with log-rank test to examine the differences in survival between patients with good and poor PS by lines of therapy. We evaluated the association between PS and risk of allcause mortality by line of therapy using a Cox proportional hazards model with adjustment for patient baseline characteristics including age at line of therapy, race, sex, index year, region, Medicare Part D LIS status, CCI, comorbid conditions not included in the CCI (dysrhythmia, other cardiac disease, anemia, osteoporosis, neutropenia, thrombocytopenia, peripheral neuropathy, venous thromboembolism, skeletalrelated events), LOS, and current line regimen. As a secondary analysis, we assessed the association of PS with mortality throughout the treatment period (up to four lines of therapy) using a marginal structural model [18] to account for time-varying
3.1. Cohort Characteristics We identified 12,547 older patients with MM who met the study inclusion criteria and initiated first-line therapy; of these, 5841 (46.6%), 2372 (18.9%), and 819 (6.5%) initiated second-, third-, and fourth-line therapy, respectively (Fig. 1). By line of therapy, proportions of patients with poor PS were 16.6% at first line, 21.8% at second line, 18.4% at third line, and 18.2% at fourth line. Compared with patients with good PS at first line, patients with poor PS were older (mean age 78.9 vs. 76.5 years), more often female (66% vs. 52%), more often African American (23% vs. 12%), more often recipients of Part D LIS (57% vs. 25%), at higher comorbidity level (CCI ≥ 5: 43% vs. 16%), more often hospitalized for ≥11 days during the baseline period (56% vs. 16%), more likely to receive monotherapy (32% vs. 19%), and less likely to receive doublet (53% vs. 61%) or triplet (10% vs. 16%) therapy. All above comparisons were statistically significant (P b 0.001). Patterns were similar for patients who advanced to second-, third-, and fourth-line therapy (Table 1).
Fig. 1. Study flow chart. MM, multiple myeloma; SCT, stem cell transplantation.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
4
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
Table 1 Demographic and clinical characteristics of study population by performance status and line of therapy. First Line
n a
Age , mean (SD), yrs. Agea, % 66–69 70–74 75–79 ≥80 Female, % Race, % White Black Other Line index yeara 2008–2009 2010 2011 2012 Census region, % Northeast Midwest South West Part D LIS, % Length of stay, days 0 1–10 ≥11 CCI, % 0–2 3–4 ≥5 Comorbidity not included in CCI, % Anemia Dysrhythmia Other cardiac diseases Osteoporosis Neutropenia Thrombocytopenia PN VTE SRE Regimen, % Monotherapy Doublets Triplets Other
Second Line
Good PS
Poor PS
10,470
2077
76.5 (6.5)
78.9 (7.2)
18.9 26.5 24.1 30.5 51.5
12.6 20.2 22.4 44.8 65.7
82.2 12.3 5.5
69.1 22.8 8.1
44.3 24.9 26.0 4.8
42.6 24.9 26.4 6.1
16.9 25.4 40.5 17.1 25.1
20.6 23.9 40.2 15.3 56.6
52.7 30.9 16.5
13.0 30.7 56.3
56.4 27.3 16.3
25.0 32.5 42.5
60.6 22.9 18.0 13.8 2.0 6.6 7.5 3.9 32.0
81.0 37.0 30.6 24.4 2.3 12.9 16.2 10.0 53.8
19.1 60.7 16.4 3.9
31.9 53.4 10.3 4.3
Third Line
Good PS
Poor PS
P
4566
1275
b0.001 b0.001
76.3 (6.1)
78.3 (6.6)
17.1 29.6 24.6 28.7 49.5
11.8 23.1 25.1 40.0 60.9
84.6 10.0 5.3
71.6 20.6 7.8
23.4 24.9 27.3 24.4
22.3 25.4 26.6 25.7
16.5 25.5 40.3 17.7 20.2
17.9 25.7 39.8 16.7 51.9
49.6 28.7 21.8
15.7 25.7 58.6
38.8 29.4 31.8
16.3 27.9 55.8
68.4 25.6 20.8 12.5 10.5 13.9 16.1 10.1 35.6
85.8 37.4 33.3 23.0 8.2 19.5 26.1 18.5 57.9
24.8 49.2 20.8 5.2
30.7 49.3 16.9 3.2
b0.001 b0.001
0.060
Good PS
Poor PS
P
1935
437
P
670
149
P
b0.001 b0.001
76.2 (5.9)
78.1 (6.2)
b0.001 b0.001
76.4 (5.8)
78.0 (5.9)
0.003 0.035
16.3 31.6 25.5 26.6 49.4
9.2 24.7 28.6 37.5 62.5
13.3 32.4 28.2 26.1 50.0
7.4 28.2 28.2 36.2 59.1
87.2 8.3 4.5
70.0 20.6 9.4
9.8 22.2 31.5 36.5
12.2 24.0 30.4 33.4
16.5 25.8 41.3 16.4 17.7
19.2 21.7 40.7 18.3 49.7
53.1 26.3 20.6
21.7 24.9 53.3
38.8 31.2 30.0
16.5 27.0 56.5
67.3 23.5 20.0 11.5 14.7 15.3 21.1 10.6 30.0
86.0 38.0 31.6 18.8 13.5 26.3 30.9 20.8 52.0
26.8 43.2 22.8 7.2
32.5 45.3 17.2 5.0
b0.001 b0.001
b0.001 b0.001
86.1 8.7 5.2
b0.001 b0.001
75.2 b
0.532 2.1 17.3 35.8 44.8
4.1 18.1 36.2 41.6
16.0 26.7 39.4 17.9 17.8
24.8 21.5 36.9 16.8 45.6
55.7 25.4 19.0
23.5 26.2 50.3
39.6 31.0 29.4
16.1 33.6 50.3
66.9 27.3 18.4 12.2 15.8 17.3 23.7 9.9 26.9
83.2 34.9 28.9 14.8 14.8 25.5 32.9 20.1 44.3
27.2 41.9 22.8 8.1
28.9 43.6 20.1 7.4
0.071
b0.001
b0.001 b0.001 b0.001 b0.001 0.510 b0.001 b0.001 b0.001 b0.001 0.007
0.045 0.002
c
0.193
b0.001
b0.001 b0.001 b0.001 b0.001 0.017 b0.001 b0.001 b0.001 b0.001 b0.001
b0.001 b0.001
0.300
0.624
b0.001
b0.001 b0.001 b0.001 b0.001 0.415 b0.001 b0.001 b0.001 b0.001 b0.001
Poor PS
0.389
b0.001
b0.001 b0.001
Fourth Line
Good PS
b0.001 b0.001
b0.001
b0.001 0.064 0.004 0.402 0.748 0.021 0.020 b0.001 b0.001 0.878
CCI, Charlson Comorbidity Index; LIS, low income subsidy; PN, peripheral neuropathy; SREs, skeletal related events; VTE, venous thromboembolism. a Defined at treatment index date of line of therapy. b Values for cells with 10 or fewer patients are suppressed. c Values suppressed to avoid deriving the values for cells with 10 or fewer patients. Table 2 Association between poor PS status and all-cause mortality by line of therapy. Adjusteda
Unadjusted Line of therapy First line Good PS Poor PS Second line Good PS Poor PS Third line Good PS Poor PS Fourth line Good PS Poor PS
Total, n
Mean (SD) Follow-up, yrs.
Death, n (%)
Mortality rate (per 100 patient-years)
HR (95% CI)
P
HR (95% CI)
P
10,470 2077
1.19 (0.98) 0.98 (0.92)
2542 (24.3) 967 (46.6)
20.4 47.6
Reference 2.26 (2.10–2.43)
b0.001
Reference 1.28 (1.18–1.40)
b0.001
4566 1275
0.97 (0.81) 0.83 (0.74)
911 (20.0) 484 (38.0)
20.6 45.6
Reference 2.17 (1.94–2.42)
b0.001
Reference 1.55 (1.36–1.77)
b0.001
1935 437
0.76 (0.66) 0.69 (0.60)
447 (23.1) 173 (40.3)
30.4 58.3
Reference 1.88 (1.58–2.25)
b0.001
Reference 1.35 (1.10–1.65)
0.004
670 149
0.66 (0.55) 0.63 (0.50)
144 (21.5) 51 (34.2)
32.8 54.1
Reference 1.60 (1.16–2.20)
0.004
Reference 1.22 (0.84–1.76)
0.29
CI, confidence interval; HR, Hazard ratio; PS, performance status; SD, standard deviation; yrs., years. a Covariates in the model included age, sex, race, index year, region, Deyo-adapted Charlson comorbidity score, dysrhythmia, other cardiac disease, anemia, osteoporosis, neutropenia, thrombocytopenia, peripheral neuropathy, venous thromboembolism, skeletal-related events, length of hospital stay, Medicare Part D low income subsidy status, and regimen.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
3.2. Mortality Outcomes by Performance Status by Lines of Therapy Median follow-up in years was shorter for patients with poor than for those with good PS at each line of therapy and decreased consistently from first-line to fourth-line therapy for patients with poor and good PS: respectively, 0.68 (inter-quartile range, 0.31–1.38) vs. 0.92 (0.45–1.62) at first line; 0.60 (0.31–1.11) vs. 0.71 (0.39–1.29) at second line; 0.52 (0.28–0.96) vs. 0.56 (0.30–1.01) at third line; 0.53 (0.28–0.80) vs. 0.48 (0.28–0.88) at fourth line. During first-line therapy, 967 (47%) patients with poor PS and 2542 (24%) patients with good PS died; corresponding all-cause mortality rates were 48 and 20 per 100 patient-years. Mortality rates were higher for patients with poor than for those with good PS consistently across second- to fourth-line therapy (Table 2). Unadjusted survival was worse for patients with poor than for those with good PS consistently across first- to fourth-line therapy (Fig. 2; P b 0.001 for first- to third-line; P = 0.004 for fourth-line). Three-year survival for patients with poor and good PS was 34% and 61% at first-line, 40% and 59% at second-line among patients who survived to receive second-line therapy, and 25% and 53% at third-line among patients who survived to receive third-line therapy, respectively. Among patients who survived to receive fourth-line therapy, 1-year survival at fourth-line was 59% and 71% for patients with poor and good PS, respectively. Compared with patients with good PS, the adjusted hazards ratio (HR) (95% CI) of mortality for patients with poor PS was 1.28 (1.18– 1.40) at first-line, 1.55 (1.36–1.77) at second-line, 1.35 (1.10–1.65) at third-line, and 1.22 (0.84–1.76) at fourth-line therapy after controlling for baseline characteristics (Table 2). 3.3. Mortality Outcomes by Performance Status Throughout Lines of Therapy
1.0 Good PS Poor PS
0.8 0.6 0.4 0.2 0.0 0
1
2
3
4
initiation of first-line therapy (median [IQR] 1.08 [0.35–2.03] vs. 1.67 [0.90–2.71]). During follow-up, 1297 (62%) patients with poor PS and 4425 (42%) patients with good PS died; corresponding mortality rates were 46 and 23 per 100 patient-years. The median overall survival (95% CI) was shorter for patients with poor than for those with good PS (1.38 [1.28–1.50] vs. 3.13 [3.02–3.25] years). The HR for mortality was 1.36 (95% CI: 1.27–1.45) for patients with poor PS compared with patients with good PS after adjusting for baseline fixed and timevarying characteristics using the weighted Cox proportional hazards model. 4. Discussion In this study, we applied a claims-based poor disability status prediction model [8;12] as a proxy measure for PS and assessed the association between PS and mortality in a population-based cohort of older US adults with MM. We found that patients with poor PS were older, had higher comorbidity levels, were more likely to be treated with monotherapy and less likely to be treated with triplets, and had worse overall survival than patients with good PS. A claims-based proxy measure for PS was associated with significantly increased risk of mortality during each of the first four lines of therapy after adjustment for baseline characteristics, indicating that PS is an independent risk factor for mortality. Two other claims-based algorithms for PS have been developed among patients with cancer. Salloum et al. [10] used claims-based service use measures, age, and an indicator of stage IV disease to predict physician-reported good PS among patients with stages II to IV lung cancer from a tumor registry. A recent study by Sheffield et al. [11] evaluated patients with cancer and commercial or Medicare supplemental insurance coverage receiving care in outpatient community oncology practices in the US. In their study, health service indicators and ageand frailty-related diagnoses were used to predict physician-reported poor PS. Both studies used PS measures as the gold standard in
Survival probability during 2LT
Survival probability during 1LT
Overall follow-up in years (line 1 through line 4) was shorter for patients with poor than for those with good PS, with PS defined at
5
1.0 Good PS Poor PS
0.8 0.6 0.4 0.2 0.0
5
0
1
2
Years from start of 1LT 4840 747
1834 271
690 93
Number at risk Good PS 4566 Poor PS 1275
204 25
1.0
Survival probability during 4LT
Survival probability during 3LT
Number at risk Good PS 10470 Poor PS 2077
Good PS Poor PS
0.8 0.6 0.4 0.2 0.0 0
1
2
3
4
5
1602 366
535 120
486 98
116 17
23 *
4
5
158 23
22 *
1.0 Good PS Poor PS
0.8 0.6 0.4 0.2 0.0 0
1
Years from start of 3LT Number at risk Good PS 1935 Poor PS 437
3
Years from start of 2LT
2
3
4
Years from start of 4LT *
Number at risk Good PS 670 Poor PS 149
138 26
20 *
*
Fig. 2. Unadjusted survival for patients with good and poor PS across first- to fourth-line therapy. LT, line of therapy; PS, performance status.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
6
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
algorithm development. However, the algorithm for PS developed by Salloum et al. was based on a small sample of 442 patients with lung cancer and included an indicator for stage IV disease. In the study by Sheffield et al., less than half of patients were aged older than 65 years with Medicare as supplementary insurance, and they may not represent the entire Medicare population. Additionally, PS measures were available for b30% of patients, who may represent a select group. Thus, the proxy measures for PS from these two studies may not be applicable to general Medicare patients with MM. Davidoff et al. developed a claims-based prediction model for disability status to proxy ECOG PS, defined as poor or good based on a dichotomous measure for the predicted probability of poor disability status in the general Medicare population [8]. In their validation study [12], the quartile of predicted probability of poor disability status was a significant independent predictor for cancer-directed treatment. ECOG PS with a 6-point scale between 0 (fully active) to 5 (dead) and Karnofsky PS with a linear scale between 0 (dead) and 100 (normally active, without evidence of disease) are the two most commonly used measures in oncology; ECOG PS is more preferred due to its simplicity and being less prone to observer error [19]. Of note, Davidoff's proxy measure of PS does not include comorbid conditions, making it possible to adjust for PS and comorbidity separately. In our study, we applied Davidoff's prediction model for poor disability status in older Medicare patients with MM as a proxy for PS and found that PS (predicted probability ≥0.11 vs. b0.11) was associated with a significantly increased risk of mortality during each line of therapy among the first four lines after adjustment for baseline characteristics; HRs (95% CI) ranged from 1.28 (1.18–1.40) during the first line to 1.55 (1.36–1.77) during the second line. The association between poor PS and increased risk of mortality is consistent with two recent studies. Mayer et al. evaluated the agreement between three measures of Medicare claims-based proxies for poor physical function among Medicare beneficiaries diagnosed with stage II/III colon cancer [20]. The study found that higher predicted probability of poor function (N 10% vs. b5%) was associated with a higher risk of mortality (adjusted HR (95% CI): 1.62 [1.53–1.70]). Casebeer et al. [21] applied the Davidoff disability prediction model to Medicare patients with metastatic non-small-cell lung cancer identified from Humana Research Database and found that patients with poor PS had an 84% higher probability of death (adjusted HR [95% CI] for PS 3–4 vs. PS 0–1: 1.84 [1.13–2.97], P = 0.014). There are several limitations of this study. First, patients with newly diagnosed MM were identified using a validated claims-based algorithm for identifying patients with MM [13]; however, some patients with prevalent MM may be included. We applied the algorithm to Medicare claims between 2007 and 2012 and selected patients with the disease index date on or after January 1, 2008, to ensure that patients included in the study had at least 1 year pre-index period without qualifying claims for MM. However, patients diagnosed with MM before January 1, 2007, or before enrolling in Medicare between 2007 and 2012, would be misidentified as newly diagnosed with MM. Since we also required patients to receive no chemotherapy, radiotherapy, stem cell transplant, or drug treatment specific to MM during the 12 months before the disease index date, this misclassification likely affected only patients with MM index dates in the earlier years of the study period. Additionally, confounding is a challenge in observational studies of medication utilization. In this study, we adjusted for important covariates that may be potential confounders in multivariable regression models. However, because some health measures, such as international staging system, cytogenetics, body mass index, smoking status, etc., are not identified in healthcare administrative databases and could not be adjusted for, some unmeasured confounding is expected. Finally, this study was performed in patients with Medicare fee-for-service coverage. MM is prevalent in the older population, and the results of our study are expected to be generalizable to the older Medicare population. However, our findings may not apply to patients who are not enrolled in Medicare Parts A, B, and/or D.
This study demonstrates that the claims-based poor disability status prediction model performed as expected when applied for defining proxy PS in older Medicare patients with MM. Patients with poor PS were older, had higher comorbidity levels, and had worse overall survival than patients with good PS. The claims-based proxy for PS was an independent risk factor for mortality. In this study, the Davidoff algorithm was used as a starting point for analysis of PS in the Medicare MM population. Further studies assessing whether this proxy measure of PS is an independent predictor for choice of treatment, and whether the association between PS and mortality is consistent by age groups and regimens, are warranted. Conflicts of Interest SL: nothing to disclose; TN: employed by OptumLabs, participating in UnitedHealth Group employee stock purchase plan; JL, consulting for Fibrogen; WW, SV, AY: Amgen, Inc. employee/stock owner; VM: consulting for Merck, GSK, Takeda; honoraria: Celgene, Genentech, Pharmacyclics, Gilead; royalties, Up-to-Date; SU: research funding from Amgen, Array Biopharma, BMS, Celgene, Janssen, Merck, Pharmacyclics, Sanofi, Takeda; consulting for Amgen, BMS, Celgene, Janssen, Merck, SkylineDx, Takeda; speaker for Amgen, Celgene, Janssen, Sanofi, Takeda. Author Contributions Substantial contributions to the conception or design of the work: SL, JL, WW, AY; or the acquisition, analysis, or interpretation of data for the work: TN, SL, JL, WW, AY. Drafting the work or revising it critically for important intellectual content: SL, JL, TN, WW, AY, SV, VM, SU. Final approval of the version to be published: all authors. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved: all authors. Acknowledgments This work was supported by Amgen Inc., Thousand Oaks, California, United States. The contract provides for the Minneapolis Medical Research Foundation authors to have final determination of manuscript content. The authors thank Chronic Disease Research Group colleagues Anne Shaw for manuscript preparation, and Nan Booth, MSW, MPH, ELS, for manuscript editing. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.jgo.2019.01.007. Reference [1] National Cancer Institute. Surveillance epidemiology and end results (SEER) program: cancer stat facts. Myeloma. Available at: https://seer.cancer.gov/statfacts/ html/mulmy.html. [Accessed November 27, 2018]. [2] Howlader N, Noone AM, Krapcho M, Miller D, Bishop K, Kosary C L, et al. (eds). SEER Cancer Statistics Review, 1975–2014, National Cancer Institute. Bethesda, MD, based on November 2016 SEER data submission, posted to the SEER web site, April 2017. Available at: https://seer.cancer.gov/archive/csr/1975_2014/. [Accessed November 27, 2018]1. [3] Kumar SK, Rajkumar SV, Dispenzieri A, Lacy MQ, Hayman SR, Buadi FK, et al. Improved survival in multiple myeloma and the impact of novel therapies. Blood 2008;111:2516–20. [4] Campagnaro EL, Goebel TE, Lazarus HM. Management of elderly patients with plasma cell myeloma. Drugs Aging 2015;32:427–42. [5] Facon T, Hulin C, Dimopoulos MA, Belch A, Meuleman N, Mohty M, et al. A frailty scale predicts outcomes in patients with newly diagnosed multiple myeloma who are ineligible for transplant treated with continuous lenalidomide plus low-dose dexamethasone in the FIRST Trial. [Abstract]. Presented at: 57th ASH Annual Meeting and Exposition December 5-8, 2015 Olrando, FL 2015.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx [6] Hope AA, Gong MN, Guerra C, Wunsch H. Frailty before critical illness and mortality for elderly medicare beneficiaries. J Am Geriatr Soc 2015;63:1121–8. [7] Zweegman S, Engelhardt M, Larocca A. Elderly patients with multiple myeloma: towards a frailty approach? Curr Opin Oncol 2017;29:315–21. [8] Davidoff AJ, Zuckerman IH, Pandya N, Hendrick F, Ke X, Hurria A, et al. A novel approach to improve health status measurement in observational claimsbased studies of cancer treatment and outcomes. J Geriatr Oncol 2013;4: 157–65. [9] Giordano SH, Kuo YF, Duan Z, Hortobagyi GN, Freeman J, Goodwin JS. Limits of observational data in determining outcomes from cancer therapy. Cancer 2008;112: 2456–66. [10] Salloum RG, Smith TJ, Jensen GA, Lafata JE. Using claims-based measures to predict performance status score in patients with lung cancer. Cancer 2011;117: 1038–48. [11] Sheffield KM, Bowman L, Smith DM, Li L, Hess LM, Montejano LB, et al. Development and validation of a claims-based approach to proxy ECOG performance status across ten tumor groups. J Comp Eff Res 2018;7:193–208. [12] Davidoff AJ, Gardner LD, Zuckerman IH, Hendrick F, Ke X, Edelman MJ. Validation of disability status, a claims-based measure of functional status for cancer treatment and outcomes studies. Med Care 2014;52:500–10. [13] Princic N, Gregory C, Willson T, Mahue M, Felici D, Werther W, et al. Development and validation of an slgorithm to identify patients with multiple myeloma using sdministrative claims data. Front Oncol 2016;6:224.
7
[14] Anderson KC, Alsina M, Atanackovic D, Biermann JS, Chandler JC, Costello C, et al. Multiple myeloma, version 2.2016: clinical practice guidelines in oncology. J Natl Compr Canc Netw 2015;13:1398–435. [15] Quan H, Sundararajan V, Halfon P, Fong A, Burnand B, Luthi JC, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005;43:1130–9. [16] Hebert PL, Geiss LS, Tierney EF, Engelgau MM, Yawn BP, McBean AM. Identifying persons with diabetes using medicare claims data. Am J Med Qual 1999;14:270–7. [17] Aly A, Onukwugha E, Woods C, Mullins CD, Kwok Y, Qian Y, et al. Measurement of skeletal related events in SEER-medicare: a comparison of claims-based methods. BMC Med Res Methodol 2015;15:65. [18] Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000;11:550–60. [19] Kelly CM, Shahrokni A. Moving beyond Karnofsky and ECOG performance status assessments with new technologies. J Oncol 2016;2016 [Article ID 6186543, 13 pages]. [20] Mayer SE, Tan HJ, Peacock Hinton S, Hester LL, Sturmer T, Faurot KR, et al. Medicare claims-based measures of poor physical functoin and associations with treatment and mortality in older colon cancer patients [Abstract]. Pharmacoepidemiol Drug Saf 2017;26. https://doi.org/10.1002/pds.4275 174, Abstract 287. [21] Casebeer A, Antol DD, DeClue RW, Hopson S, Li Y, Khoury R, et al. The relationship between guideline-recommended initiation of therapy, outcomes, and cost for patients with metastatic non-small cell lung cancer. J Manag Care Spec Pharm 2018; 24:554–64.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
Contents lists available at ScienceDirect
Journal of Geriatric Oncology
Mortality by a proxy performance status as defined by a claims-based measure for disability status in older patients with newly diagnosed multiple myeloma in the United States Shuling Li a,⁎, Tanya Natwick a,1, Jiannong Liu a, Vicki A. Morrison a,b, Sarah Vidito c, Winifred Werther d, Akeem A. Yusuf c, Saad Z. Usmani e a
Chronic Disease Research Group, Hennepin Healthcare Research Institute, 701 Park Avenue, Suite S4.100, Minneapolis, MN 55415, USA Hennepin Healthcare Systems, 701 Park Avenue, Minneapolis, MN 55415, USA c Amgen Inc, 1 Amgen Center Dr, Thousand Oaks, California 91320, USA d Amgen Inc, 1120 Veterans Blvd, South San Francisco, California 94-80, USA e Carolinas HealthCare System, 1021 Morehead Medical Drive, Charlotte, North Carolina 28204, USA b
a r t i c l e
i n f o
Article history: Received 2 August 2018 Received in revised form 20 December 2018 Accepted 8 January 2019 Available online xxxx Keywords: Administrative claims Cancer Medicare Mortality Population-based cohort Proxy for performance status
a b s t r a c t Objectives: We applied a claims-based definition of disability status as a proxy for performance status (PS) and examined associations between PS and mortality in a population-based cohort of older US adults with multiple myeloma (MM). Materials and Methods: We identified older (≥66 years) Medicare beneficiaries diagnosed with MM January 1, 2008-December 31, 2011, who began first-line therapy in the study period (through December 31, 2012). We estimated predicted probability of poor PS for each patient at initiation of each line up to fourth-line therapy, classified as poor (predicted probability ≥0.11) or good (b0.11) PS, and examined mortality. Crude overall survival was estimated using the Kaplan-Meier method with log-rank test for survival comparison between PS groups. Cox proportional hazards models evaluated the association between poor PS and mortality risk, adjusted for baseline characteristics by lines of therapy. Results: Of 12,547 patients, 5841, 2372, and 819 initiated second-, third-, and fourth-line in the study period. Poor PS proportions were 16.6%, 21.8%, 18.4%, and 18.2% at each line. Crude overall survival was worse for poor PS patients across lines (P b 0.01 for each). Adjusted hazards ratios (95% CI) of mortality for patients with poor versus good PS were 1.28 (1.18–1.40), first-line; 1.55 (1.36–1.77), second-line; 1.35 (1.10–1.65), third-line; 1.22 (0.84– 1.76), fourth-line. Conclusion: The claims-based prediction model for disability status performed as expected as a proxy for PS in older Medicare patients with MM. PS was an independent risk factor for mortality. Further studies assessing the effect of PS on mortality by therapies are warranted. © 2019 Published by Elsevier Ltd.
1. Introduction Multiple myeloma (MM) is a malignancy of predominantly older patients, accounting for 1.8% of all incident cancer cases and 2.1% of all cancer deaths in the US population [1]. In 2018, an estimated 30,770 new cases were diagnosed, and 12,770 patients died of MM [1]. MM
⁎ Corresponding author at: Chronic Disease Research Group, Hennepin Healthcare Research Institute, 701 Park Avenue, Suite S2.100, Minneapolis, MN 55404, USA. E-mail addresses: [email protected] (S. Li), [email protected] (J. Liu), [email protected] (W. Werther), [email protected] (A.A. Yusuf), [email protected] (S.Z. Usmani). 1 Currently affiliated with OptumLabs, Minneapolis, Minnesota, USA.
incidence increases with age, with N60% of diagnoses and 78% of deaths occurring in patients aged 65 years or older [1,2]. Over the past decade, several novel MM therapies have been approved, leading to substantial improvement in survival [3]. Older patients, however, benefit less from new treatments, possibly due to decreased treatment tolerance, increased health burden, and worse biology [4]. Patients with poor health status are especially at risk of treatment-related adverse events and poor clinical outcomes due to reduced physiologic reserve and limited adaptability to stressful events [5–7]. Performance status ([PS], e.g., Eastern Cooperative Oncology Group [ECOG] or Karnofsky Performance Status), one dimension of patients' health status, is defined by ability to carry out physical activities, some of which are related to self-care and independent living [8]. PS is used
https://doi.org/10.1016/j.jgo.2019.01.007 1879-4068/© 2019 Published by Elsevier Ltd.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
2
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
by clinicians to evaluate patients' health condition and assess risks and benefits in treatment choices. This information is usually not available in administrative claims data. Thus, observational studies using medical claims data to evaluate treatment patterns and compare the effectiveness of therapies on survival are often subject to bias due to unmeasured confounding, such as PS [9]. Although two claims-based PS algorithms have been developed using data from a single tumor registry for lung cancer [10] and from outpatient community oncology practices including ten tumor types [11], a claims-based definition of PS for older general Medicare patients with MM has not been established. Davidoff and colleagues developed an algorithm linking self-reported functional status measures based on activities of daily living (ADL) and instrumental ADL to a disability status scale (0–4, representing a progression of increasing degrees of limitation) using the Medicare Current Beneficiary Survey data linked with Medicare claims. This disability status scale was a proxy for the ECOG PS scale, and was dichotomized to focus on good (ECOG 0–2) versus poor (ECOG 3–4) PS. Davidoff et al. then identified potential claims-based predictors of poor PS and developed a prediction model for proxy PS, defined as poor or good based on a dichotomous measure for the predicted probability of poor PS in the general Medicare population [8;12]. The purpose of this study was to apply the claims-based measure of disability status developed by Davidoff et al. to a population-based cohort of older US adults with MM as a proxy for PS, and to assess the association between PS and mortality. 2. Materials and Methods 2.1. Data Source The Centers for Medicare & Medicaid Services database for the 100% sample of Medicare beneficiaries with hematologic cancers including MM from 2007 to 2012 was used in this study. The files include demographics, Part D low-income subsidy (LIS) status, Medicare enrollment status, dates of service, and diagnostic and procedure codes. Mortality information (date of all-cause death) was obtained using the date of death field from the Master Beneficiary Summary File. We applied to and received approval from the Human Subjects Research Committee of Hennepin County Medical Center/Hennepin Healthcare System, Inc. 2.2. Study design and Population This was a retrospective cohort study covering January 1, 2007, through December 31, 2012. The study population included Medicare beneficiaries diagnosed with MM between January 1, 2008, and December 31, 2011, based on a validated algorithm using a combination of International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes for MM (203.00, 203.01, and 203.02) and diagnostic tests or treatment [13]. The algorithm required patients to have an MM diagnosis (index diagnosis) and (i) three additional MM diagnoses during the 90 days before the index diagnosis AND either one bone marrow or two other diagnostic tests during the 90 days before the index diagnosis (Appendix A1) or (ii) chemotherapy within 180 days after the index diagnosis (Appendix A2-A4). The index diagnosis date was identified as the disease index date. We further required patients to be aged at least 66 years at the disease index date, to have initiated a first-line therapy following the disease index date (the date of treatment initiation defined as the first-line index date), and to be continuously enrolled in fee-for-service Medicare Parts A, B, and D from 12 months before the disease index date to the first-line index date. In an effort to include only patients with newly diagnosed MM, we excluded patients who had received any chemotherapy, radiotherapy, or stem cell transplant (Appendix B, D), or drug treatments specific to MM (Appendix C) in the 12 months before the disease index date. Patients with missing census region based on the state of residence were also excluded.
For each treated patient, first-line to fourth-line therapies were identified based on an algorithm derived from healthcare administrative claims and electronic medical record databases. The first-line therapy started on the first date of any MM-related chemotherapy. Appendix C provides a list of codes for drugs available during the study period. The first line ended at the earliest of the following: 1) a gap of N90 days in all MM-related therapies included for the first-line therapy; 2) addition of or switch to a new MM therapy (not including prednisone) that was not adjunctive to the first-line regimens; 3) death; or 4) disenrollment from any Medicare Part A, B, or D coverage or December 31, 2012. Second-, third-, and fourth-line therapies were defined using the same approach. Drug regimens were identified based on the anti-myeloma medications used within 90 days of the line of therapy index date using National Drug Codes from Medicare Part D prescription drug event claims and Healthcare Common Procedure Coding System codes from Part B line items and Part A outpatient claims (Appendix C) and classified as monotherapy, doublets, or triplets based on the National Comprehensive Cancer Network MM treatment guidelines [14]. Regimens not identified as one of these three types were classified as “other.” Prednisone alone was not considered monotherapy for MM, but was allowed to be part of a drug regimen. The baseline period for each line of therapy was the 12 months before the treatment index date of the line, during which PS, comorbid conditions, Deyo-adapted Charlson Comorbidity Index (CCI) [15], length of hospital stay (LOS), and LIS were defined. The follow-up period for each line of therapy was the corresponding line duration, and the overall follow-up period was from the first-line start date to the fourth-line end date. 2.3. Measures A claims-based poor disability status prediction model developed and validated by Davidoff et al. [8;12] was used to estimate the probability of poor disability status as a proxy measure for PS at line-oftherapy initiation for each patient via the following steps: First, we defined the healthcare service predictors as described by Davidoff et al. [8] from Medicare claims during the baseline period for each line-of-therapy cohort. Healthcare service predictors included evaluation and management visits, other visit types, minor or ambulatory procedures, major procedures, preventive services, durable medical equipment use, imaging, and other service indicators. The prediction model also included Medicaid enrollment status and sex, identified from the Master Beneficiary Summary file. Next, we applied the regression coefficients provided in Appendix C from Davidoff et al. [8] to the set of constructed measures for each patient to generate a predicted probability of poor disability status, with high values representing high probability. Finally, based on the predicted probability, patients were classified as poor PS (probability of poor disability status ≥0.11) or good PS (b0.11). The 0.11 cut-off was chosen by Davidoff et al. to maximize the best balance of sensitivity and specificity among a cohort of Medicare beneficiaries [8]. We performed a spline curve analysis of effect of predicted probability on patient survival and found no evidence to reject the use of this cut-off to identify patients with poor disability status used as proxy for poor PS in a cohort of older patients with MM. Demographics (date of birth, sex, race, geographic region) were identified in the Medicare Beneficiary Summary file with age calculated at each line-of-therapy index date and categorized into four groups (66–69, 70–74, 75–79, ≥80 years). Medicare Part D LIS was identified in the Medicare beneficiary enrollment file. Hospital LOS during the baseline period was defined from inpatient claims and categorized into three groups (0, 1–10, ≥11 days). Level of comorbidity was assessed during the baseline period using the Quan version of the CCI [15], and categorized into four groups (0, 1–2, 3–4, ≥5). Comorbid conditions not included in the CCI were also identified during the baseline period. Most comorbid conditions were defined using standard methodology
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
[16], and skeletal-related events were identified using the algorithm of “base case” definition from Aly et al. [17] All-cause mortality was assessed during the relevant follow-up period for each line of therapy cohort and overall throughout first-line to fourth-line therapy.
3
poor PS and time-varying confounders (regimens and comorbid conditions) updated at the subsequent line of therapy index date, since these factors may potentially predict PS and mortality and serve as intermediate variables on the causal pathway between poor PS and death.
3. Results
2.4. Statistical Analyses Baseline characteristics are described by PS and lines of therapy. We report counts and percentages for categorical variables, mean (standard deviation [SD]) and median (interquartile range [IQR]) for continuous variables. Mortality rates are expressed as numbers of patients who died per 100 patient-years. Crude all-cause survival was estimated using the Kaplan-Meier method with log-rank test to examine the differences in survival between patients with good and poor PS by lines of therapy. We evaluated the association between PS and risk of allcause mortality by line of therapy using a Cox proportional hazards model with adjustment for patient baseline characteristics including age at line of therapy, race, sex, index year, region, Medicare Part D LIS status, CCI, comorbid conditions not included in the CCI (dysrhythmia, other cardiac disease, anemia, osteoporosis, neutropenia, thrombocytopenia, peripheral neuropathy, venous thromboembolism, skeletalrelated events), LOS, and current line regimen. As a secondary analysis, we assessed the association of PS with mortality throughout the treatment period (up to four lines of therapy) using a marginal structural model [18] to account for time-varying
3.1. Cohort Characteristics We identified 12,547 older patients with MM who met the study inclusion criteria and initiated first-line therapy; of these, 5841 (46.6%), 2372 (18.9%), and 819 (6.5%) initiated second-, third-, and fourth-line therapy, respectively (Fig. 1). By line of therapy, proportions of patients with poor PS were 16.6% at first line, 21.8% at second line, 18.4% at third line, and 18.2% at fourth line. Compared with patients with good PS at first line, patients with poor PS were older (mean age 78.9 vs. 76.5 years), more often female (66% vs. 52%), more often African American (23% vs. 12%), more often recipients of Part D LIS (57% vs. 25%), at higher comorbidity level (CCI ≥ 5: 43% vs. 16%), more often hospitalized for ≥11 days during the baseline period (56% vs. 16%), more likely to receive monotherapy (32% vs. 19%), and less likely to receive doublet (53% vs. 61%) or triplet (10% vs. 16%) therapy. All above comparisons were statistically significant (P b 0.001). Patterns were similar for patients who advanced to second-, third-, and fourth-line therapy (Table 1).
Fig. 1. Study flow chart. MM, multiple myeloma; SCT, stem cell transplantation.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
4
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
Table 1 Demographic and clinical characteristics of study population by performance status and line of therapy. First Line
n a
Age , mean (SD), yrs. Agea, % 66–69 70–74 75–79 ≥80 Female, % Race, % White Black Other Line index yeara 2008–2009 2010 2011 2012 Census region, % Northeast Midwest South West Part D LIS, % Length of stay, days 0 1–10 ≥11 CCI, % 0–2 3–4 ≥5 Comorbidity not included in CCI, % Anemia Dysrhythmia Other cardiac diseases Osteoporosis Neutropenia Thrombocytopenia PN VTE SRE Regimen, % Monotherapy Doublets Triplets Other
Second Line
Good PS
Poor PS
10,470
2077
76.5 (6.5)
78.9 (7.2)
18.9 26.5 24.1 30.5 51.5
12.6 20.2 22.4 44.8 65.7
82.2 12.3 5.5
69.1 22.8 8.1
44.3 24.9 26.0 4.8
42.6 24.9 26.4 6.1
16.9 25.4 40.5 17.1 25.1
20.6 23.9 40.2 15.3 56.6
52.7 30.9 16.5
13.0 30.7 56.3
56.4 27.3 16.3
25.0 32.5 42.5
60.6 22.9 18.0 13.8 2.0 6.6 7.5 3.9 32.0
81.0 37.0 30.6 24.4 2.3 12.9 16.2 10.0 53.8
19.1 60.7 16.4 3.9
31.9 53.4 10.3 4.3
Third Line
Good PS
Poor PS
P
4566
1275
b0.001 b0.001
76.3 (6.1)
78.3 (6.6)
17.1 29.6 24.6 28.7 49.5
11.8 23.1 25.1 40.0 60.9
84.6 10.0 5.3
71.6 20.6 7.8
23.4 24.9 27.3 24.4
22.3 25.4 26.6 25.7
16.5 25.5 40.3 17.7 20.2
17.9 25.7 39.8 16.7 51.9
49.6 28.7 21.8
15.7 25.7 58.6
38.8 29.4 31.8
16.3 27.9 55.8
68.4 25.6 20.8 12.5 10.5 13.9 16.1 10.1 35.6
85.8 37.4 33.3 23.0 8.2 19.5 26.1 18.5 57.9
24.8 49.2 20.8 5.2
30.7 49.3 16.9 3.2
b0.001 b0.001
0.060
Good PS
Poor PS
P
1935
437
P
670
149
P
b0.001 b0.001
76.2 (5.9)
78.1 (6.2)
b0.001 b0.001
76.4 (5.8)
78.0 (5.9)
0.003 0.035
16.3 31.6 25.5 26.6 49.4
9.2 24.7 28.6 37.5 62.5
13.3 32.4 28.2 26.1 50.0
7.4 28.2 28.2 36.2 59.1
87.2 8.3 4.5
70.0 20.6 9.4
9.8 22.2 31.5 36.5
12.2 24.0 30.4 33.4
16.5 25.8 41.3 16.4 17.7
19.2 21.7 40.7 18.3 49.7
53.1 26.3 20.6
21.7 24.9 53.3
38.8 31.2 30.0
16.5 27.0 56.5
67.3 23.5 20.0 11.5 14.7 15.3 21.1 10.6 30.0
86.0 38.0 31.6 18.8 13.5 26.3 30.9 20.8 52.0
26.8 43.2 22.8 7.2
32.5 45.3 17.2 5.0
b0.001 b0.001
b0.001 b0.001
86.1 8.7 5.2
b0.001 b0.001
75.2 b
0.532 2.1 17.3 35.8 44.8
4.1 18.1 36.2 41.6
16.0 26.7 39.4 17.9 17.8
24.8 21.5 36.9 16.8 45.6
55.7 25.4 19.0
23.5 26.2 50.3
39.6 31.0 29.4
16.1 33.6 50.3
66.9 27.3 18.4 12.2 15.8 17.3 23.7 9.9 26.9
83.2 34.9 28.9 14.8 14.8 25.5 32.9 20.1 44.3
27.2 41.9 22.8 8.1
28.9 43.6 20.1 7.4
0.071
b0.001
b0.001 b0.001 b0.001 b0.001 0.510 b0.001 b0.001 b0.001 b0.001 0.007
0.045 0.002
c
0.193
b0.001
b0.001 b0.001 b0.001 b0.001 0.017 b0.001 b0.001 b0.001 b0.001 b0.001
b0.001 b0.001
0.300
0.624
b0.001
b0.001 b0.001 b0.001 b0.001 0.415 b0.001 b0.001 b0.001 b0.001 b0.001
Poor PS
0.389
b0.001
b0.001 b0.001
Fourth Line
Good PS
b0.001 b0.001
b0.001
b0.001 0.064 0.004 0.402 0.748 0.021 0.020 b0.001 b0.001 0.878
CCI, Charlson Comorbidity Index; LIS, low income subsidy; PN, peripheral neuropathy; SREs, skeletal related events; VTE, venous thromboembolism. a Defined at treatment index date of line of therapy. b Values for cells with 10 or fewer patients are suppressed. c Values suppressed to avoid deriving the values for cells with 10 or fewer patients. Table 2 Association between poor PS status and all-cause mortality by line of therapy. Adjusteda
Unadjusted Line of therapy First line Good PS Poor PS Second line Good PS Poor PS Third line Good PS Poor PS Fourth line Good PS Poor PS
Total, n
Mean (SD) Follow-up, yrs.
Death, n (%)
Mortality rate (per 100 patient-years)
HR (95% CI)
P
HR (95% CI)
P
10,470 2077
1.19 (0.98) 0.98 (0.92)
2542 (24.3) 967 (46.6)
20.4 47.6
Reference 2.26 (2.10–2.43)
b0.001
Reference 1.28 (1.18–1.40)
b0.001
4566 1275
0.97 (0.81) 0.83 (0.74)
911 (20.0) 484 (38.0)
20.6 45.6
Reference 2.17 (1.94–2.42)
b0.001
Reference 1.55 (1.36–1.77)
b0.001
1935 437
0.76 (0.66) 0.69 (0.60)
447 (23.1) 173 (40.3)
30.4 58.3
Reference 1.88 (1.58–2.25)
b0.001
Reference 1.35 (1.10–1.65)
0.004
670 149
0.66 (0.55) 0.63 (0.50)
144 (21.5) 51 (34.2)
32.8 54.1
Reference 1.60 (1.16–2.20)
0.004
Reference 1.22 (0.84–1.76)
0.29
CI, confidence interval; HR, Hazard ratio; PS, performance status; SD, standard deviation; yrs., years. a Covariates in the model included age, sex, race, index year, region, Deyo-adapted Charlson comorbidity score, dysrhythmia, other cardiac disease, anemia, osteoporosis, neutropenia, thrombocytopenia, peripheral neuropathy, venous thromboembolism, skeletal-related events, length of hospital stay, Medicare Part D low income subsidy status, and regimen.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
3.2. Mortality Outcomes by Performance Status by Lines of Therapy Median follow-up in years was shorter for patients with poor than for those with good PS at each line of therapy and decreased consistently from first-line to fourth-line therapy for patients with poor and good PS: respectively, 0.68 (inter-quartile range, 0.31–1.38) vs. 0.92 (0.45–1.62) at first line; 0.60 (0.31–1.11) vs. 0.71 (0.39–1.29) at second line; 0.52 (0.28–0.96) vs. 0.56 (0.30–1.01) at third line; 0.53 (0.28–0.80) vs. 0.48 (0.28–0.88) at fourth line. During first-line therapy, 967 (47%) patients with poor PS and 2542 (24%) patients with good PS died; corresponding all-cause mortality rates were 48 and 20 per 100 patient-years. Mortality rates were higher for patients with poor than for those with good PS consistently across second- to fourth-line therapy (Table 2). Unadjusted survival was worse for patients with poor than for those with good PS consistently across first- to fourth-line therapy (Fig. 2; P b 0.001 for first- to third-line; P = 0.004 for fourth-line). Three-year survival for patients with poor and good PS was 34% and 61% at first-line, 40% and 59% at second-line among patients who survived to receive second-line therapy, and 25% and 53% at third-line among patients who survived to receive third-line therapy, respectively. Among patients who survived to receive fourth-line therapy, 1-year survival at fourth-line was 59% and 71% for patients with poor and good PS, respectively. Compared with patients with good PS, the adjusted hazards ratio (HR) (95% CI) of mortality for patients with poor PS was 1.28 (1.18– 1.40) at first-line, 1.55 (1.36–1.77) at second-line, 1.35 (1.10–1.65) at third-line, and 1.22 (0.84–1.76) at fourth-line therapy after controlling for baseline characteristics (Table 2). 3.3. Mortality Outcomes by Performance Status Throughout Lines of Therapy
1.0 Good PS Poor PS
0.8 0.6 0.4 0.2 0.0 0
1
2
3
4
initiation of first-line therapy (median [IQR] 1.08 [0.35–2.03] vs. 1.67 [0.90–2.71]). During follow-up, 1297 (62%) patients with poor PS and 4425 (42%) patients with good PS died; corresponding mortality rates were 46 and 23 per 100 patient-years. The median overall survival (95% CI) was shorter for patients with poor than for those with good PS (1.38 [1.28–1.50] vs. 3.13 [3.02–3.25] years). The HR for mortality was 1.36 (95% CI: 1.27–1.45) for patients with poor PS compared with patients with good PS after adjusting for baseline fixed and timevarying characteristics using the weighted Cox proportional hazards model. 4. Discussion In this study, we applied a claims-based poor disability status prediction model [8;12] as a proxy measure for PS and assessed the association between PS and mortality in a population-based cohort of older US adults with MM. We found that patients with poor PS were older, had higher comorbidity levels, were more likely to be treated with monotherapy and less likely to be treated with triplets, and had worse overall survival than patients with good PS. A claims-based proxy measure for PS was associated with significantly increased risk of mortality during each of the first four lines of therapy after adjustment for baseline characteristics, indicating that PS is an independent risk factor for mortality. Two other claims-based algorithms for PS have been developed among patients with cancer. Salloum et al. [10] used claims-based service use measures, age, and an indicator of stage IV disease to predict physician-reported good PS among patients with stages II to IV lung cancer from a tumor registry. A recent study by Sheffield et al. [11] evaluated patients with cancer and commercial or Medicare supplemental insurance coverage receiving care in outpatient community oncology practices in the US. In their study, health service indicators and ageand frailty-related diagnoses were used to predict physician-reported poor PS. Both studies used PS measures as the gold standard in
Survival probability during 2LT
Survival probability during 1LT
Overall follow-up in years (line 1 through line 4) was shorter for patients with poor than for those with good PS, with PS defined at
5
1.0 Good PS Poor PS
0.8 0.6 0.4 0.2 0.0
5
0
1
2
Years from start of 1LT 4840 747
1834 271
690 93
Number at risk Good PS 4566 Poor PS 1275
204 25
1.0
Survival probability during 4LT
Survival probability during 3LT
Number at risk Good PS 10470 Poor PS 2077
Good PS Poor PS
0.8 0.6 0.4 0.2 0.0 0
1
2
3
4
5
1602 366
535 120
486 98
116 17
23 *
4
5
158 23
22 *
1.0 Good PS Poor PS
0.8 0.6 0.4 0.2 0.0 0
1
Years from start of 3LT Number at risk Good PS 1935 Poor PS 437
3
Years from start of 2LT
2
3
4
Years from start of 4LT *
Number at risk Good PS 670 Poor PS 149
138 26
20 *
*
Fig. 2. Unadjusted survival for patients with good and poor PS across first- to fourth-line therapy. LT, line of therapy; PS, performance status.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
6
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx
algorithm development. However, the algorithm for PS developed by Salloum et al. was based on a small sample of 442 patients with lung cancer and included an indicator for stage IV disease. In the study by Sheffield et al., less than half of patients were aged older than 65 years with Medicare as supplementary insurance, and they may not represent the entire Medicare population. Additionally, PS measures were available for b30% of patients, who may represent a select group. Thus, the proxy measures for PS from these two studies may not be applicable to general Medicare patients with MM. Davidoff et al. developed a claims-based prediction model for disability status to proxy ECOG PS, defined as poor or good based on a dichotomous measure for the predicted probability of poor disability status in the general Medicare population [8]. In their validation study [12], the quartile of predicted probability of poor disability status was a significant independent predictor for cancer-directed treatment. ECOG PS with a 6-point scale between 0 (fully active) to 5 (dead) and Karnofsky PS with a linear scale between 0 (dead) and 100 (normally active, without evidence of disease) are the two most commonly used measures in oncology; ECOG PS is more preferred due to its simplicity and being less prone to observer error [19]. Of note, Davidoff's proxy measure of PS does not include comorbid conditions, making it possible to adjust for PS and comorbidity separately. In our study, we applied Davidoff's prediction model for poor disability status in older Medicare patients with MM as a proxy for PS and found that PS (predicted probability ≥0.11 vs. b0.11) was associated with a significantly increased risk of mortality during each line of therapy among the first four lines after adjustment for baseline characteristics; HRs (95% CI) ranged from 1.28 (1.18–1.40) during the first line to 1.55 (1.36–1.77) during the second line. The association between poor PS and increased risk of mortality is consistent with two recent studies. Mayer et al. evaluated the agreement between three measures of Medicare claims-based proxies for poor physical function among Medicare beneficiaries diagnosed with stage II/III colon cancer [20]. The study found that higher predicted probability of poor function (N 10% vs. b5%) was associated with a higher risk of mortality (adjusted HR (95% CI): 1.62 [1.53–1.70]). Casebeer et al. [21] applied the Davidoff disability prediction model to Medicare patients with metastatic non-small-cell lung cancer identified from Humana Research Database and found that patients with poor PS had an 84% higher probability of death (adjusted HR [95% CI] for PS 3–4 vs. PS 0–1: 1.84 [1.13–2.97], P = 0.014). There are several limitations of this study. First, patients with newly diagnosed MM were identified using a validated claims-based algorithm for identifying patients with MM [13]; however, some patients with prevalent MM may be included. We applied the algorithm to Medicare claims between 2007 and 2012 and selected patients with the disease index date on or after January 1, 2008, to ensure that patients included in the study had at least 1 year pre-index period without qualifying claims for MM. However, patients diagnosed with MM before January 1, 2007, or before enrolling in Medicare between 2007 and 2012, would be misidentified as newly diagnosed with MM. Since we also required patients to receive no chemotherapy, radiotherapy, stem cell transplant, or drug treatment specific to MM during the 12 months before the disease index date, this misclassification likely affected only patients with MM index dates in the earlier years of the study period. Additionally, confounding is a challenge in observational studies of medication utilization. In this study, we adjusted for important covariates that may be potential confounders in multivariable regression models. However, because some health measures, such as international staging system, cytogenetics, body mass index, smoking status, etc., are not identified in healthcare administrative databases and could not be adjusted for, some unmeasured confounding is expected. Finally, this study was performed in patients with Medicare fee-for-service coverage. MM is prevalent in the older population, and the results of our study are expected to be generalizable to the older Medicare population. However, our findings may not apply to patients who are not enrolled in Medicare Parts A, B, and/or D.
This study demonstrates that the claims-based poor disability status prediction model performed as expected when applied for defining proxy PS in older Medicare patients with MM. Patients with poor PS were older, had higher comorbidity levels, and had worse overall survival than patients with good PS. The claims-based proxy for PS was an independent risk factor for mortality. In this study, the Davidoff algorithm was used as a starting point for analysis of PS in the Medicare MM population. Further studies assessing whether this proxy measure of PS is an independent predictor for choice of treatment, and whether the association between PS and mortality is consistent by age groups and regimens, are warranted. Conflicts of Interest SL: nothing to disclose; TN: employed by OptumLabs, participating in UnitedHealth Group employee stock purchase plan; JL, consulting for Fibrogen; WW, SV, AY: Amgen, Inc. employee/stock owner; VM: consulting for Merck, GSK, Takeda; honoraria: Celgene, Genentech, Pharmacyclics, Gilead; royalties, Up-to-Date; SU: research funding from Amgen, Array Biopharma, BMS, Celgene, Janssen, Merck, Pharmacyclics, Sanofi, Takeda; consulting for Amgen, BMS, Celgene, Janssen, Merck, SkylineDx, Takeda; speaker for Amgen, Celgene, Janssen, Sanofi, Takeda. Author Contributions Substantial contributions to the conception or design of the work: SL, JL, WW, AY; or the acquisition, analysis, or interpretation of data for the work: TN, SL, JL, WW, AY. Drafting the work or revising it critically for important intellectual content: SL, JL, TN, WW, AY, SV, VM, SU. Final approval of the version to be published: all authors. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved: all authors. Acknowledgments This work was supported by Amgen Inc., Thousand Oaks, California, United States. The contract provides for the Minneapolis Medical Research Foundation authors to have final determination of manuscript content. The authors thank Chronic Disease Research Group colleagues Anne Shaw for manuscript preparation, and Nan Booth, MSW, MPH, ELS, for manuscript editing. Appendix A. Supplementary data Supplementary data to this article can be found online at https://doi. org/10.1016/j.jgo.2019.01.007. Reference [1] National Cancer Institute. Surveillance epidemiology and end results (SEER) program: cancer stat facts. Myeloma. Available at: https://seer.cancer.gov/statfacts/ html/mulmy.html. [Accessed November 27, 2018]. [2] Howlader N, Noone AM, Krapcho M, Miller D, Bishop K, Kosary C L, et al. (eds). SEER Cancer Statistics Review, 1975–2014, National Cancer Institute. Bethesda, MD, based on November 2016 SEER data submission, posted to the SEER web site, April 2017. Available at: https://seer.cancer.gov/archive/csr/1975_2014/. [Accessed November 27, 2018]1. [3] Kumar SK, Rajkumar SV, Dispenzieri A, Lacy MQ, Hayman SR, Buadi FK, et al. Improved survival in multiple myeloma and the impact of novel therapies. Blood 2008;111:2516–20. [4] Campagnaro EL, Goebel TE, Lazarus HM. Management of elderly patients with plasma cell myeloma. Drugs Aging 2015;32:427–42. [5] Facon T, Hulin C, Dimopoulos MA, Belch A, Meuleman N, Mohty M, et al. A frailty scale predicts outcomes in patients with newly diagnosed multiple myeloma who are ineligible for transplant treated with continuous lenalidomide plus low-dose dexamethasone in the FIRST Trial. [Abstract]. Presented at: 57th ASH Annual Meeting and Exposition December 5-8, 2015 Olrando, FL 2015.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007
S. Li et al. / Journal of Geriatric Oncology xxx (2019) xxx [6] Hope AA, Gong MN, Guerra C, Wunsch H. Frailty before critical illness and mortality for elderly medicare beneficiaries. J Am Geriatr Soc 2015;63:1121–8. [7] Zweegman S, Engelhardt M, Larocca A. Elderly patients with multiple myeloma: towards a frailty approach? Curr Opin Oncol 2017;29:315–21. [8] Davidoff AJ, Zuckerman IH, Pandya N, Hendrick F, Ke X, Hurria A, et al. A novel approach to improve health status measurement in observational claimsbased studies of cancer treatment and outcomes. J Geriatr Oncol 2013;4: 157–65. [9] Giordano SH, Kuo YF, Duan Z, Hortobagyi GN, Freeman J, Goodwin JS. Limits of observational data in determining outcomes from cancer therapy. Cancer 2008;112: 2456–66. [10] Salloum RG, Smith TJ, Jensen GA, Lafata JE. Using claims-based measures to predict performance status score in patients with lung cancer. Cancer 2011;117: 1038–48. [11] Sheffield KM, Bowman L, Smith DM, Li L, Hess LM, Montejano LB, et al. Development and validation of a claims-based approach to proxy ECOG performance status across ten tumor groups. J Comp Eff Res 2018;7:193–208. [12] Davidoff AJ, Gardner LD, Zuckerman IH, Hendrick F, Ke X, Edelman MJ. Validation of disability status, a claims-based measure of functional status for cancer treatment and outcomes studies. Med Care 2014;52:500–10. [13] Princic N, Gregory C, Willson T, Mahue M, Felici D, Werther W, et al. Development and validation of an slgorithm to identify patients with multiple myeloma using sdministrative claims data. Front Oncol 2016;6:224.
7
[14] Anderson KC, Alsina M, Atanackovic D, Biermann JS, Chandler JC, Costello C, et al. Multiple myeloma, version 2.2016: clinical practice guidelines in oncology. J Natl Compr Canc Netw 2015;13:1398–435. [15] Quan H, Sundararajan V, Halfon P, Fong A, Burnand B, Luthi JC, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005;43:1130–9. [16] Hebert PL, Geiss LS, Tierney EF, Engelgau MM, Yawn BP, McBean AM. Identifying persons with diabetes using medicare claims data. Am J Med Qual 1999;14:270–7. [17] Aly A, Onukwugha E, Woods C, Mullins CD, Kwok Y, Qian Y, et al. Measurement of skeletal related events in SEER-medicare: a comparison of claims-based methods. BMC Med Res Methodol 2015;15:65. [18] Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000;11:550–60. [19] Kelly CM, Shahrokni A. Moving beyond Karnofsky and ECOG performance status assessments with new technologies. J Oncol 2016;2016 [Article ID 6186543, 13 pages]. [20] Mayer SE, Tan HJ, Peacock Hinton S, Hester LL, Sturmer T, Faurot KR, et al. Medicare claims-based measures of poor physical functoin and associations with treatment and mortality in older colon cancer patients [Abstract]. Pharmacoepidemiol Drug Saf 2017;26. https://doi.org/10.1002/pds.4275 174, Abstract 287. [21] Casebeer A, Antol DD, DeClue RW, Hopson S, Li Y, Khoury R, et al. The relationship between guideline-recommended initiation of therapy, outcomes, and cost for patients with metastatic non-small cell lung cancer. J Manag Care Spec Pharm 2018; 24:554–64.
Please cite this article as: S. Li, T. Natwick, J. Liu, et al., Mortality by a proxy performance status as defined by a claims-based measure for disability status i..., J Geriatr Oncol, https://doi.org/10.1016/j.jgo.2019.01.007