Research Article For reprint orders, please contact: [email protected]
Comparative effectiveness of follow-up imaging approaches in pancreatic cancer
Aim: Although PET imaging is sometimes used in follow-up of pancreatic cancer, evidence regarding comparative effectiveness of PET and older imaging modalities is limited. Patients & methods: Linked cancer registry and Medicare claims data were analyzed to examine patterns of imaging and effects on treatment patterns and survival among newly diagnosed pancreatic cancer patients from 2003 to 2007. Results: 12% of patients received PET during follow-up. In a time-varying exposure model, computed tomography/MRI was associated with lower mortality risk relative to PET in surgical patients (HR: 0.66; 95% CI: 0.52–0.83). In a subset analysis, type of follow-up imaging before 180 days was not associated with mortality after 180 days (computed tomography/MRI vs PET; hazard ratio: 0.98; 95% CI: 0.84–1.16). Conclusion: Follow-up PET is uncommon among Medicare beneficiaries with pancreatic cancer, and is generally used late in the disease course. This pattern of PET use was not associated with decreased mortality risk compared with conventional imaging. Keywords: comparative effectiveness • imaging • pancreatic cancer • PET
Pancreatic cancer is a common and highly lethal solid tumor, affecting more than 45,000 patients in the USA and resulting in more than 38,000 deaths in 2013 [1] . The clinical course of pancreatic cancer is relentlessly aggressive, with most presentations of advanced stage, for which median survival is 30 days was present, the prior course of chemotherapy was considered ended and the next subsequent chemotherapy claim marked the beginning of a new course. The number of days on chemotherapy was adjusted for days surviving from diagnosis, to account for the fact that patients with longer survival have greater opportunity to receive treatment. To examine radiation treatment duration, we counted the number of days during follow-up on which the patient had a claim for radiation services and adjusted for number of days surviving. Potential covariates were chosen based on known prognostic factors in pancreatic cancer and factors influencing access to care. Demographic variables included age, sex, race, marital status and Medicaid coverage. comorbidities were determined from the Medicare data using the Klabunde modification of the Charlson Comorbidity Index [15] . County-level data on education, income, urban/rural designation and percent uninsured were obtained from the ARF. The ARF was also used to obtain data on the number of hospitals with CT, MRI and PET scanners and oncologic services. To account for possible stage migration due to the increased sensitivity of PET in detecting metastases at diagnosis [16] , we controlled for type of diagnostic imaging, defined as a claim for PET, CT or MRI within 45 days before or after diagnosis date.
Models for adjusted treatment duration outcome
Analytic approach
Early-exposure model for mortality outcome
To examine the association between PET exposure and receipt of chemotherapy and radiation treatment, we conducted cross-sectional analyses using negative binomial models. To examine the association between PET exposure and mortality we used two analytic approaches: a time-varying exposure model and an early-exposure model, described in more detail below. Considering our sample size and to develop parsimonious models, we selected covariates by first conducting univariate analyses to identify variables associated with either the exposure or the outcome of interest. These variables were included in the initial model. Final models were fitted in which the variables were reduced on the basis of statistical significance and clinical concern for potential confounding. The effect of treatment on outcomes was examined by preplanned stratification of models by treatment pathway (surgical, borderline or metastatic). All analyses for this paper were generated using SAS/STAT software, Version 9.2 (TS2M3) of the SAS System for Windows on the XP_PRO platform (SAS Institute Inc., NC, USA). The models are described individually below.
Analyses of exposure/outcome relationships can also be biased by defining the exposure (imaging) in such a way that the outcome (death) cannot possibly occur during some portion of the outcome assessment period, a problem commonly called immortal time bias [17] . To address immortal time bias, we constructed a second early-exposure model which segregated the time of exposure from the time of outcome assessment. For this analysis, we limited our sample to patients surviving 180 days postdiagnosis (n = 2010). Patients were then categorized according to the imaging received between days 45 and 180 as ‘PET’ (with or without other imaging), ‘CT/MRI’ (only CT and/or MRI) or ‘never imaged’. Survival time after 180 days was determined. Cox proportional hazard models were estimated for the entire sample and stratified by treatment group.
J. Comp. Eff. Res. (2014) 3(5)
We utilized negative binomial models to examine the relationship between PET exposure and chemotherapy or radiation duration adjusted for days surviving. The samples for these analyses were limited to individuals who received the treatment of interest (n = 2984 for chemotherapy and n = 1049 for radiation). Individuals who received both chemotherapy and radiation were included in both models. Time-varying exposure model for mortality outcome
In order to account for possible misattribution of survival time to PET scans that occurred after a significant portion of the survival time had already accrued, we employed a time-varying exposure model approach for our whole-cohort analysis (Figure 2) . This approach accounted for the survival time attained by the patient before and after each type of imaging occurred. Patients were initially placed in the ‘no scan’ category. Survival time was then attributed to the ‘no scan’ category until receipt of other imaging. After the first CT or MRI, survival time was attributed to the ‘CT/MRI’ category. After exposure to PET, all subsequent survival time was attributed to the ‘PET’ category. Cox proportional hazard models were estimated for the entire sample and stratified by treatment group (surgery, borderline or metastatic). Patients with unknown stage were excluded.
Results Sample characteristics
Sample characteristics are shown in Table 1. The average age at diagnosis was 77 years. 21% of patients were nonwhite, including 10% African-Americans, 3% Hispanics, 5% Asians and 3% others. Due to
future science group
Comparative effectiveness of follow-up imaging approaches in pancreatic cancer
Research Article
Patient receiving only CT: Survival attributed to no scan
Diagnosis
Survival attributed to CT
First CT
Death
Patient receiving CT followed by PET:
Survival attributed to no scan
Diagnosis
Survival attributed to CT
First CT
Survival attributed to PET
First PET
Death
Patient receiving PET followed by CT: Survival attributed to no scan
Diagnosis
Survival attributed to PET
First PET
Survival attributed to PET
First CT
Death
Figure 2. Diagram illustrates the time-varying exposure model of the effect of imaging modality on survival. For all patients, survival time was initially assigned to a ‘no scan’ group. For patients with claims for CT only, all survival time subsequent to the first claim for CT was attributed to the ‘CT’ category. For patients with claims for CT and subsequent PET, survival time after the first claim for CT was attributed to the ‘CT’ category until the first claim for PET. All survival time after first claim for PET was attributed to the ‘PET’ category. CT: Computed tomography.
small sample sizes, race was collapsed to a dichotomous variable. Approximately 10% of patients were diagnosed with localized disease, 32% with regional disease and 42% with metastatic disease. 16% of patients were unstaged, a proportion consistent with prior findings from SEER registry data [18] . Most patients died d uring the study period (n = 4465, 96%). 12% of patients (n = 578) underwent at least one PET or PET/CT during their pancreatic cancer follow-up. 97% of these patients (n = 571) also received a distinct CT/MRI during follow-up. Patients receiving PET were generally younger, healthier and resided in higher income areas than other groups. PET usage increased over time. Among patients who received PET, median time between diagnosis and first PET scan was 197 days, and 25% of patients received their first PET scan at >354 days postdiagnosis.
and among those treated, PET patients had more days on chemotherapy and radiation (Table 1) . In adjusted negative binomial models (data not shown), survival-adjusted chemotherapy treatment duration for patients receiving CT/MRI or never imaged was lower than that of patients receiving PET (hazard ratio [HR]: 0.82; 95% CI: 0.72–0.93 and HR: 0.26; 95% CI: 0.22–0.31, respectively). Chemotherapy duration was positively associated with younger age, and negatively associated with Medicaid eligibility and absence of any diagnostic imaging. Radiation treatment duration for patients receiving CT/MRI was not significantly different from that for patients receiving PET (HR: 0.96; 95% CI: 0.85–1.08), but adjusted duration for patients with no follow-up imaging was significantly lower (HR: 0.59; 95% CI: 0.48–0.73). No other significant predictors of radiation treatment were found in multivariate analysis.
Association of imaging use and adjusted treatment duration
Time-varying exposure model
In unadjusted analysis, patients receiving PET were more likely to receive chemotherapy and radiation,
future science group
After adjustment for relevant covariates, receipt of conventional follow-up imaging was associated with lower mortality risk relative to receipt of PET
www.futuremedicine.com
495
Research Article Reeder-Hayes, Freburger, Feaganes et al.
Table 1. Sample characteristics overall and by follow-up imaging group (n = 4652). Covariate
Overall; n (%)
PET (n = 578); n (%)
CT/MRI (n = 2409); n (%)
Never imaged (n = 1665); n (%)
Age (mean = 77, range = 66–103); years: – 66–69 – 70–74 – 75–79 – 80+
792 (17) 1057 (23) 1105 (24) 1698 (36)
138 (24) 152 (26) 166 (29) 122 (21)
470 (19) 619 (26) 577 (24) 743 (31)
184 (11) 286 (17) 362 (22) 833 (50)
Race: – White – Non-white
3674 (79) 978 (21)
475 (82) 103 (18)
1885 (78) 524 (22)
1314 (79) 351 (21)
Sex: – Male
2003 (43)
266 (46)
1070 (44)
667 (40)
Marital status: – Married – Unmarried – Unknown
2320 (50) 1986 (43) 346 (7)
358 (62) 186 (32) 34 (6)
1277 (53) 953 (40) 179 (7)
685 (41) 847 (51) 132 (8)
% uninsured: – Lower (20%)
3336 (72) 1316 (28)
398 (69) 180 (31)
1701 (71) 708 (29)
1237 (74) 428 (26)
Income (median household): – Low ($60K)
962 (21) 1835 (39) 1024 (22) 831 (18)
70 (12) 235 (41) 131 (23) 142 (25)
479 (20) 971 (40) 532 (22) 427 (18)
413 (25) 629 (38) 361 (27) 262 (16)
Comorbidity index: –0 –1 – 2+
1965 (42) 1438 (31) 1198 (26)
273 (47) 174 (30) 129 (22)
1033 (43) 761 (32) 599 (25)
659 (40) 503 (31) 470 (29)
Region of USA: – CA – UT – NC
3018 (65) 217 (5) 1417 (30)
426 (74) 13 (2) 139 (24)
1555 (65) 108 (4) 746 (31)
1037 (62) 96 (6) 532 (32)
Year of diagnosis: – 2003 – 2004 – 2005 – 2006 – 2007
864 (18) 963 (21) 923 (20) 974 (21) 928 (20)
47 (8) 93 (16) 122 (21) 155 (27) 161 (28)
476 (20) 502 (21) 492 (20) 471 (20) 468 (19)
341 (20) 368 (22) 309 (19) 348 (21) 299 (18)
Initial treatment pathway: – Surgery – Borderline – Metastatic – Unknown
500 (11) 1477 (32) 1946 (42) 729 (16)
162 (28) 201 (35) 180 (31) 35 (6)
310 (13) 768 (32) 993 (41) 338 (14)
28 (2) 508 (31) 773 (46) 356 (21)
p-value†
Comparative effectiveness of follow-up imaging approaches in pancreatic cancer
Aim: Although PET imaging is sometimes used in follow-up of pancreatic cancer, evidence regarding comparative effectiveness of PET and older imaging modalities is limited. Patients & methods: Linked cancer registry and Medicare claims data were analyzed to examine patterns of imaging and effects on treatment patterns and survival among newly diagnosed pancreatic cancer patients from 2003 to 2007. Results: 12% of patients received PET during follow-up. In a time-varying exposure model, computed tomography/MRI was associated with lower mortality risk relative to PET in surgical patients (HR: 0.66; 95% CI: 0.52–0.83). In a subset analysis, type of follow-up imaging before 180 days was not associated with mortality after 180 days (computed tomography/MRI vs PET; hazard ratio: 0.98; 95% CI: 0.84–1.16). Conclusion: Follow-up PET is uncommon among Medicare beneficiaries with pancreatic cancer, and is generally used late in the disease course. This pattern of PET use was not associated with decreased mortality risk compared with conventional imaging. Keywords: comparative effectiveness • imaging • pancreatic cancer • PET
Pancreatic cancer is a common and highly lethal solid tumor, affecting more than 45,000 patients in the USA and resulting in more than 38,000 deaths in 2013 [1] . The clinical course of pancreatic cancer is relentlessly aggressive, with most presentations of advanced stage, for which median survival is 30 days was present, the prior course of chemotherapy was considered ended and the next subsequent chemotherapy claim marked the beginning of a new course. The number of days on chemotherapy was adjusted for days surviving from diagnosis, to account for the fact that patients with longer survival have greater opportunity to receive treatment. To examine radiation treatment duration, we counted the number of days during follow-up on which the patient had a claim for radiation services and adjusted for number of days surviving. Potential covariates were chosen based on known prognostic factors in pancreatic cancer and factors influencing access to care. Demographic variables included age, sex, race, marital status and Medicaid coverage. comorbidities were determined from the Medicare data using the Klabunde modification of the Charlson Comorbidity Index [15] . County-level data on education, income, urban/rural designation and percent uninsured were obtained from the ARF. The ARF was also used to obtain data on the number of hospitals with CT, MRI and PET scanners and oncologic services. To account for possible stage migration due to the increased sensitivity of PET in detecting metastases at diagnosis [16] , we controlled for type of diagnostic imaging, defined as a claim for PET, CT or MRI within 45 days before or after diagnosis date.
Models for adjusted treatment duration outcome
Analytic approach
Early-exposure model for mortality outcome
To examine the association between PET exposure and receipt of chemotherapy and radiation treatment, we conducted cross-sectional analyses using negative binomial models. To examine the association between PET exposure and mortality we used two analytic approaches: a time-varying exposure model and an early-exposure model, described in more detail below. Considering our sample size and to develop parsimonious models, we selected covariates by first conducting univariate analyses to identify variables associated with either the exposure or the outcome of interest. These variables were included in the initial model. Final models were fitted in which the variables were reduced on the basis of statistical significance and clinical concern for potential confounding. The effect of treatment on outcomes was examined by preplanned stratification of models by treatment pathway (surgical, borderline or metastatic). All analyses for this paper were generated using SAS/STAT software, Version 9.2 (TS2M3) of the SAS System for Windows on the XP_PRO platform (SAS Institute Inc., NC, USA). The models are described individually below.
Analyses of exposure/outcome relationships can also be biased by defining the exposure (imaging) in such a way that the outcome (death) cannot possibly occur during some portion of the outcome assessment period, a problem commonly called immortal time bias [17] . To address immortal time bias, we constructed a second early-exposure model which segregated the time of exposure from the time of outcome assessment. For this analysis, we limited our sample to patients surviving 180 days postdiagnosis (n = 2010). Patients were then categorized according to the imaging received between days 45 and 180 as ‘PET’ (with or without other imaging), ‘CT/MRI’ (only CT and/or MRI) or ‘never imaged’. Survival time after 180 days was determined. Cox proportional hazard models were estimated for the entire sample and stratified by treatment group.
J. Comp. Eff. Res. (2014) 3(5)
We utilized negative binomial models to examine the relationship between PET exposure and chemotherapy or radiation duration adjusted for days surviving. The samples for these analyses were limited to individuals who received the treatment of interest (n = 2984 for chemotherapy and n = 1049 for radiation). Individuals who received both chemotherapy and radiation were included in both models. Time-varying exposure model for mortality outcome
In order to account for possible misattribution of survival time to PET scans that occurred after a significant portion of the survival time had already accrued, we employed a time-varying exposure model approach for our whole-cohort analysis (Figure 2) . This approach accounted for the survival time attained by the patient before and after each type of imaging occurred. Patients were initially placed in the ‘no scan’ category. Survival time was then attributed to the ‘no scan’ category until receipt of other imaging. After the first CT or MRI, survival time was attributed to the ‘CT/MRI’ category. After exposure to PET, all subsequent survival time was attributed to the ‘PET’ category. Cox proportional hazard models were estimated for the entire sample and stratified by treatment group (surgery, borderline or metastatic). Patients with unknown stage were excluded.
Results Sample characteristics
Sample characteristics are shown in Table 1. The average age at diagnosis was 77 years. 21% of patients were nonwhite, including 10% African-Americans, 3% Hispanics, 5% Asians and 3% others. Due to
future science group
Comparative effectiveness of follow-up imaging approaches in pancreatic cancer
Research Article
Patient receiving only CT: Survival attributed to no scan
Diagnosis
Survival attributed to CT
First CT
Death
Patient receiving CT followed by PET:
Survival attributed to no scan
Diagnosis
Survival attributed to CT
First CT
Survival attributed to PET
First PET
Death
Patient receiving PET followed by CT: Survival attributed to no scan
Diagnosis
Survival attributed to PET
First PET
Survival attributed to PET
First CT
Death
Figure 2. Diagram illustrates the time-varying exposure model of the effect of imaging modality on survival. For all patients, survival time was initially assigned to a ‘no scan’ group. For patients with claims for CT only, all survival time subsequent to the first claim for CT was attributed to the ‘CT’ category. For patients with claims for CT and subsequent PET, survival time after the first claim for CT was attributed to the ‘CT’ category until the first claim for PET. All survival time after first claim for PET was attributed to the ‘PET’ category. CT: Computed tomography.
small sample sizes, race was collapsed to a dichotomous variable. Approximately 10% of patients were diagnosed with localized disease, 32% with regional disease and 42% with metastatic disease. 16% of patients were unstaged, a proportion consistent with prior findings from SEER registry data [18] . Most patients died d uring the study period (n = 4465, 96%). 12% of patients (n = 578) underwent at least one PET or PET/CT during their pancreatic cancer follow-up. 97% of these patients (n = 571) also received a distinct CT/MRI during follow-up. Patients receiving PET were generally younger, healthier and resided in higher income areas than other groups. PET usage increased over time. Among patients who received PET, median time between diagnosis and first PET scan was 197 days, and 25% of patients received their first PET scan at >354 days postdiagnosis.
and among those treated, PET patients had more days on chemotherapy and radiation (Table 1) . In adjusted negative binomial models (data not shown), survival-adjusted chemotherapy treatment duration for patients receiving CT/MRI or never imaged was lower than that of patients receiving PET (hazard ratio [HR]: 0.82; 95% CI: 0.72–0.93 and HR: 0.26; 95% CI: 0.22–0.31, respectively). Chemotherapy duration was positively associated with younger age, and negatively associated with Medicaid eligibility and absence of any diagnostic imaging. Radiation treatment duration for patients receiving CT/MRI was not significantly different from that for patients receiving PET (HR: 0.96; 95% CI: 0.85–1.08), but adjusted duration for patients with no follow-up imaging was significantly lower (HR: 0.59; 95% CI: 0.48–0.73). No other significant predictors of radiation treatment were found in multivariate analysis.
Association of imaging use and adjusted treatment duration
Time-varying exposure model
In unadjusted analysis, patients receiving PET were more likely to receive chemotherapy and radiation,
future science group
After adjustment for relevant covariates, receipt of conventional follow-up imaging was associated with lower mortality risk relative to receipt of PET
www.futuremedicine.com
495
Research Article Reeder-Hayes, Freburger, Feaganes et al.
Table 1. Sample characteristics overall and by follow-up imaging group (n = 4652). Covariate
Overall; n (%)
PET (n = 578); n (%)
CT/MRI (n = 2409); n (%)
Never imaged (n = 1665); n (%)
Age (mean = 77, range = 66–103); years: – 66–69 – 70–74 – 75–79 – 80+
792 (17) 1057 (23) 1105 (24) 1698 (36)
138 (24) 152 (26) 166 (29) 122 (21)
470 (19) 619 (26) 577 (24) 743 (31)
184 (11) 286 (17) 362 (22) 833 (50)
Race: – White – Non-white
3674 (79) 978 (21)
475 (82) 103 (18)
1885 (78) 524 (22)
1314 (79) 351 (21)
Sex: – Male
2003 (43)
266 (46)
1070 (44)
667 (40)
Marital status: – Married – Unmarried – Unknown
2320 (50) 1986 (43) 346 (7)
358 (62) 186 (32) 34 (6)
1277 (53) 953 (40) 179 (7)
685 (41) 847 (51) 132 (8)
% uninsured: – Lower (20%)
3336 (72) 1316 (28)
398 (69) 180 (31)
1701 (71) 708 (29)
1237 (74) 428 (26)
Income (median household): – Low ($60K)
962 (21) 1835 (39) 1024 (22) 831 (18)
70 (12) 235 (41) 131 (23) 142 (25)
479 (20) 971 (40) 532 (22) 427 (18)
413 (25) 629 (38) 361 (27) 262 (16)
Comorbidity index: –0 –1 – 2+
1965 (42) 1438 (31) 1198 (26)
273 (47) 174 (30) 129 (22)
1033 (43) 761 (32) 599 (25)
659 (40) 503 (31) 470 (29)
Region of USA: – CA – UT – NC
3018 (65) 217 (5) 1417 (30)
426 (74) 13 (2) 139 (24)
1555 (65) 108 (4) 746 (31)
1037 (62) 96 (6) 532 (32)
Year of diagnosis: – 2003 – 2004 – 2005 – 2006 – 2007
864 (18) 963 (21) 923 (20) 974 (21) 928 (20)
47 (8) 93 (16) 122 (21) 155 (27) 161 (28)
476 (20) 502 (21) 492 (20) 471 (20) 468 (19)
341 (20) 368 (22) 309 (19) 348 (21) 299 (18)
Initial treatment pathway: – Surgery – Borderline – Metastatic – Unknown
500 (11) 1477 (32) 1946 (42) 729 (16)
162 (28) 201 (35) 180 (31) 35 (6)
310 (13) 768 (32) 993 (41) 338 (14)
28 (2) 508 (31) 773 (46) 356 (21)
p-value†
