Health-Care Utilization by Prognosis Profile in a Managed Care Setting: Using the Surveillance, Epidemiology and End Results Cancer Survival Calculator SEER*CSC

Abstract

Background

Accurate estimation of the probability of dying of cancer versus other causes is needed to inform goals of care for cancer patients. Further, prognosis may also influence health-care utilization. This paper describes health service utilization patterns of subgroups of prostate cancer and colorectal cancer (CRC) patients with different relative probabilities of dying of their cancer or other conditions.

Methods

A retrospective cohort of cancer patients from Kaiser Permanente Colorado were divided into three groups using the predicted probabilities of dying of cancer and other causes calculated by the nomograms in the National Cancer Institute Surveillance, Epidemiology and End Results Cancer Survival Calculator. Demographic, disease-related characteristics, and health service utilization patterns were described across subgroups.

Results

The cohort consisted of 2092 patients (1102 prostate cancer and 990 CRC). A new diagnosis of cancer increased utilization of cancer-related services with rates as high as 9.1/1000 person-days for prostate cancer and 36.2/1000 person-days for CRC. Little change was observed in the number of primary and other specialty care visits from prediagnosis to 1 and 2 years postdiagnosis.

Conclusions

We found that although a new diagnosis of cancer increased utilization of cancer-related services for an extended time period, the timing of cancer diagnosis did not appear to affect other types of utilization. Future research should assess the reason for the lack of impact of cancer and unrelated comorbid conditions on utilization and whether desired outcomes of care were achieved.

Although thoughtful consideration of both estimates of risk of dying of cancer and other causes can and should inform goals of care, there are inconsistencies in how care of other cause morbidities are managed in the face of incident cancer diagnosis (1,2). For example, individuals with incident poor prognosis cancer may continue to receive statin medication for cardiovascular risk management (3). The combination of risk of dying of cancer and other causes may also influence the frequency and type of health-care services patients will utilize following diagnosis although the impact of these competing conditions may be difficult to predict. For example, patients who are undergoing cancer treatment may be less likely to receive recommended care across a number of chronic comorbid conditions (4). The impact of cancer treatment on primary care visits is unclear, with conflicting findings from prior studies (4–7). Furthermore, with the rapid uptake of early detection technologies and the aging of the population, many newly diagnosed cancer patients have a higher risk of dying of other causes than their cancer. Accurate information on cancer and other cause-related prognosis can inform medical choices for both clinicians and their patients.

The Surveillance, Epidemiology and End Results Cancer Survival Calculator (SEER*CSC) (previously called the Cancer Survival Query System), a web-based cancer prognostic tool developed for colorectal and prostate cancer by the Surveillance Research Program of the National Cancer Institute (NCI), was designed to support clinicians’ decision making with cancer and other cause prognosis in mind using data from NCI’s SEER cancer registries; Medicare claims data linked to SEER data; and Medicare claims data for noncancer patients in SEER areas (8). The nomograms were developed to predict the risk of death from cancer and other causes based on a profile of prognostic factors providing guidance to cancer specialists and generalists on the appropriateness of treatment both of the cancer and other conditions, taking into account the overall prognostic picture. More information about the SEER*CSC and underlying nomograms are provided in Feuer et al. (8) and in another article in this issue (9).

To better inform shared decision making in the face of incident cancer, we conducted two related analyses. First, we conducted an external validation of the nomograms of the SEER*CSC with a cohort of prostate cancer and colorectal cancer (CRC) patients from an integrated health system (ie, the Kaiser Permanente Colorado [KPCO] health maintenance organization) to assess how well the SEER*CSC nomograms generated estimates using a population-based dataset apply to a population of health-care patients (reference Feuer et al. in this issue). In his article, we describe findings from an analysis that used these now validated nomograms to describe the health service utilization patterns of subgroups of prostate cancer and CRC patients with different relative probabilities of dying of their cancer or other conditions. We focused specifically on the number of cancer and other cause-related specialty outpatient visits and primary care visits to describe 1) the rate of primary care utilization after diagnosis and 2) primary care visit rates were based on cancer prognosis.

Data Sources and Methods

Sample

Our sample was a retrospective cohort of members of KPCO diagnosed with incident prostate cancer or CRC between 2001 and 2008. All cases were required to have a minimum of 1 year membership in the health plan before diagnosis. Patients included were also required to have nonmissing staging information and to have been 40 years or older at the time of diagnosis.

We obtained information on cancer diagnosis, characteristics, and treatment from KPCO’s tumor registry; additional data were extracted from KPCO’s administrative databases. Racial categories were based on Kaiser Permanente’s Utility for Care Data Analysis’ Geographically Enriched Member Sociodemographics, a census-based datamart (10).

Individuals were defined as being “low” socioeconomic status (SES) if at least 20% of residents within the residence block (based on Geographical Information System and census data) had household incomes below the federal poverty level or at least 25% of residents aged 25 or older in the census block had less than a high school education.

Mortality data were ascertained using data from KPCO’s Tumor Registry, health plan administrative clinical and administrative data sources, and the State of Colorado. These data include cause of death and were used as categories of alive, died of cancer, and died of other cause based on underlying cause of death. Survival time was calculated as the number of months between the date of diagnosis and either date of death or the end of our observation period (April 30, 2012).

The comorbidity score was the sum of number of comorbidities used as inputs to the SEER*CSC algorithm: AIDS, acute myocardial infarction, old acute myocardial infarction, congestive heart failure, cerebrovascular disease, peripheral vascular disease, chronic obstructive pulmonary disease, renal disease, rheumatologic disease, ulcer disease, mild liver disease, moderate-to-severe liver disease, diabetes, dementia, and paralysis (11).

Identifying and Describing Subgroups of Cancer Patients

We used the validated nomograms for prostate cancer and CRC to create subgroups of patients based on their relative probabilities of dying of cancer and other causes. We then described the demographic and disease-related characteristics and explored health-care utilization patterns of these subgroups.

The SEER*CSC nomograms for prostate cancer and CRC and individual-level patient characteristics from the above-described KPCO clinical and administrative data on our cohort were used to calculate one- through five-year predicted probabilities of dying of cancer and other causes as a function of stage and other tumor characteristics, demographic characteristics (age, race–gender, and marital status), and comorbidity (P _cancer + P _other + P _alive = 1.0). Predicted survival was imputed if a case was missing race, marital status, or both. If a case was missing race, then a weighted average of race-specific survival curves was computed with the weights derived from Geographically Enriched Member Sociodemographics. If marital status or race and marital status were missing, the weights were derived from the cases in the KPCO dataset where race and/or marital status were available.

We used five-year probabilities of dying of cancer and from other causes to divide our population into three groups for each of prostate cancer and CRC: 1) probability of dying of cancer exceeded the probability of dying of other causes by at least 5 percentage points (ie, high cancer), 2) probability of dying of other causes exceeded the probability of dying of cancer by at least 5 percentage points (ie, high comorbidity), and 3) probabilities of dying of cancer and dying of other causes were within ±5 percentage points of each other (ie, equal).

We described rates per 1000 person-days of utilization (ie, alive and enrolled) in the year before cancer diagnosis and in each of the 2 years following diagnosis. We assessed rates of primary care visits, oncology visits, and other specialty care visits for both prostate cancer and CRC patients. For prostate cancer cases, we looked at urology visits separately, and for CRC cases, we separated surgery and gastroenterology specialty visits from other specialty care visits. Other specialty care visits (ie, other than oncology and urology for prostate cancer, and oncology, surgery, and gastroenterology for CRC) are less likely to be related to initial cancer-related treatment.

Results

A total of 2092 patients met the study criteria: 1102 with prostate cancer and 990 with CRC. The average age of prostate cancer patients was 67 (standard deviation [SD] = 8.8), and mean follow-up time was 71 (SD = 35.2) months; 259 (24%) died during the study time period. Most (n = 908; 82%) of prostate cancer patients were diagnosed at stage I and 100 (9%) were diagnosed at stage IV. Slightly more than half of CRC cancer patients were women (50.2%), average age was 68 (SD = 11.2), and mean follow-up time was 53 (SD = 35.8) months. Most CRC patients were diagnosed at stage I (n = 257, 26%); 169 (17%) were diagnosed at stage IV. Close to half (n = 458; 46%) patients died. Summary characteristics of prostate cancer and CRC patients by probability group are provided in Tables 1 and 2.

Table 1.

Characteristics by probability subgroups: five-year probability, prostate cancer*

Characteristics	High cancer: P(cancer) > P(other) (n = 81, 22%)		Equal: |P(cancer) − P(other)| < 0.05 (n = 439, 40%)		High other: P(other) > P(cancer) (n = 582, 53%)		P
	N	%	N	%	N	%
Race
White	57	70.4	357	81.3	438	75.3
Black	8	9.9	15	3.4	71	12.2
Other	0	0.0	14	3.2	4	0.7
Missing	16	19.8	53	12.1	69	11.9	<.0001
Low SES	11	13.6	43	9.8	78	13.4	.4579
Marital status
Married	43	53.1	333	75.9	407	69.9
Not married	14	17.3	74	16.9	102	17.5
Unknown	24	29.6	32	7.3	73	12.5	<.0001
Vital status
Alive	19	23.5	394	89.8	430	73.9
Died of cancer	47	58.0	26	5.9	55	9.5
Died of other	15	18.5	19	4.3	97	16.7	<.0001
Staging category
Pure stage	81	100.0	292	66.5	459	78.9
Pathological stage	0	0.0	147	33.5	123	21.1	<.0001
Gleason score
Well/moderately differentiated	7	8.6	166	37.8	352	60.5
Poorly/undifferentiated	74	91.4	273	62.2	230	39.5	<.0001
Clinical stage
Missing	0	0.0	147	33.5	123	21.1
T1a N0 M0	0	0.0	5	1.1	23	4.0
T1b N0 M0	0	0.0	2	0.5	6	1.0
T1c N0 M0	0	0.0	82	18.7	200	34.4
T2 N0 M0	0	0.0	191	43.5	220	37.8
T3 N0 M0	5	6.2	3	0.7	3	0.5
T4 N0 M0	1	1.2	0	0.0	1	0.2
T1–T3 N1 M0	3	3.7	6	1.4	1	0.2
T4 N1 M0	2	2.5	0	0.0	0	0.0
M1 (distant mets)	70	86.4	3	0.7	5	0.9	<.0001
Pathologic stage
Missing	81	100.0	292	66.5	459	78.9
pT2 pN0 pM0	0	0.0	90	20.5	99	17.0
pT3 pN0 pM0	0	0.0	51	11.6	24	4.1
pT4 pN0 pM0	0	0.0	5	1.1	0	0.0
pT2/pT3 pN1 pM0	0	0.0	1	0.2	0	0.0	<.0001
Surgery
Yes	3	3.7	154	35.1	155	26.6
No	78	96.3	285	64.9	426	73.2
Unknown	0	0.0	0	0.0	1	0.2	<.0001
Radiation
Yes	21	25.9	228	51.9	237	40.7
No	60	74.1	211	48.1	343	58.9
Unknown	0	0.0	0	0.0	2	0.3	<.0001
Hormone therapy
Yes	72	88.9	151	34.4	170	29.2
No	9	11.1	286	65.2	407	69.9
Unknown	0	0.0	2	0.5	5	0.9	<.0001
	Mean (SD)	Median (5%, 95%)	Mean (SD)	Median (5%, 95%)	Mean (SD)	Median (5%, 95%)	P
Age at diagnosis	69.4 (9.5)	68 (55, 85)	61.9 (7.9)	62 (49, 73)	70.0 (7.5)	71 (59, 82)	<.0001
Survival time, m	38.0 (25.2)	33 (8, 87)	73.6 (33.9)	71 (13, 131)	72.8 (35.1)	69 (14, 130)	<.0001
Comorbidity score	0.42 (0.65)	0 (0, 2)	0.24 (0.57)	0 (0, 1)	0.59 (0.85)	0 (0, 2)	<.0001

* SD = standard deviation; SES = socioeconomic status.

Table 2.

Characteristics by probability subgroup: 5-year probability, colorectal cancer*

Characteristics	High cancer: P(cancer) > P(other) (n = 511, 52%)		Equal: |P(cancer) − P(other)| < 0.05 (n = 277, 28%)		High other: P(other) > P(cancer) (n = 202, 20%)		P
	N	%	N	%	N	%
Male gender	259	50.7	135	48.7	99	49.0	.8454
Race
White	359	70.3	207	74.7	143	70.8
Black	21	4.1	10	3.6	5	2.5
Other	14	2.7	7	2.5	2	1.0
Missing	117	22.9	53	19.1	52	25.7	.4291
Low SES	64	12.5	38	13.7	47	23.3	.0047
Marital status
Married	257	50.3	185	66.8	87	43.1
Not married	111	21.7	59	21.3	58	28.7
Unknown	143	28.0	33	11.9	57	28.2	<.0001
Vital status
Alive	213	41.7	229	82.7	90	44.6
Died of cancer	247	48.3	23	8.3	46	22.8
Died of other	51	10.0	25	9.0	66	32.7	<.0001
Subsite
Proximal	213	41.7	137	49.5	130	64.4
Distal	131	25.6	81	29.2	40	19.8
Rectum	167	32.7	59	21.3	32	15.8	<.0001
Tumor grade
Low grade	333	65.2	238	85.9	162	80.2
High grade	178	34.8	39	14.1	40	19.8	<.0001
AJCC stage
Stage 0	4	0.8	7	2.5	7	3.5
Stage I no surgery	0	0.0	1	0.4	2	1.0
Stage I postsurgery	10	2.0	150	54.2	93	46.0
Stage IIA–B	87	17.0	102	36.8	83	41.1
Stage IIIA–C	241	47.2	17	6.1	17	8.4
Stage IV	169	33.1	0	0.0	0	0.0	<.0001
Surgery (yes)	464	90.8	276	99.6	196	97.0	<.0001
Radiation
Yes	111	21.7	18	6.5	10	5.0
No	399	78.1	259	93.5	192	95.1
Unknown	1	0.2	0	0.0	0	0.0	<.0001
Chemotherapy (yes)	355	69.5	36	13.0	18	8.9	<.0001
	Mean (SD)	Median (5%, 95%)	Mean (SD)	Median (5%, 95%)	Mean (SD)	Median (5%, 95%)	P
Age at diagnosis	65.7 (11.5)	66 (47, 94)	65.6 (9.2)	66 (50, 80)	78.1 (6.3)	78 (67, 90)	<.0001
Survival time, m	46.1 (36.3)	42 (2, 116)	67.9 (33.4)	66 (8, 123)	51.2 (31.7)	49.5 (4, 106)	<.0001
Comorbidity score	0.46 (0.76)	0 (0, 2)	0.41 (0.81)	0 (0, 2)	1.53 (1.35)	1 (0, 4)	<.0001

* AJCC = American Joint Committee on Cancer; SD = standard deviation; SES = socioeconomic status.

Descriptive Characteristics of Prostate Cancer Patients

For prostate cancer patients, the largest proportion of patients (53%) was categorized as having greater probability of dying of other causes of death than their cancer within the 5 years after diagnosis (“high other”). This group was followed by the group of patients with close to equal probability of dying of cancer and other causes (“equal”) (40%) and the group with higher probability of dying of cancer than their other causes (“high cancer”) (22%). Distribution of the characteristics (eg, treatment, age, and vital status) of the prostate cancer patients across probability groups is summarized in Table 1.

Descriptive Characteristics of Colorectal Cancer Patients

When categorizing the patients across probability groups (“high cancer,” “high other,” “equal”), we found that for CRC patients, the largest proportion of patients (52%) were categorized in the “high cancer” group. There was roughly the same proportion of patients categorized as “equal” and “high other” (28% and 20%, respectively). Details regarding the characteristics of each subgroup are provided in Table 2.

Health-Care Utilization for Prostate Cancer Patients

Health-care utilization patterns across probability groups are depicted in Figure 1 (prostate cancer) and Figure 2 (CRC patients) in visits per 1000 person-days.

Figure 1.

Visits per 1000 person-days by five-year probability of death from cancer/other cause for prostate cancer (urology, oncology, specialty, and primary care).

Figure 2.

Visits per 1000 person-days by five-year probability of death from cancer/other cause for colorectal cancer (oncology, specialty, and primary care).

For prostate cancer patients, we included both oncology and urology visits separately to represent care that is more likely associated with initial cancer treatment. The three groups had similar rates of urology visits the year after diagnosis (11–12 per 1000 person-days) but for oncology visits, the “high cancer” group had substantially more visits the year after diagnosis (9.1/1000 person-days) than the other two groups (1.8 and 1.4 for “equal” and “high other” groups, respectively). Primary care visits were approximately the same for the “high cancer” (9.5/1000 person-days) and “high other” groups (8.8/1000 person-days), and the “equal” group had a slightly lower rate (6.4 visits/1000 person-days) during the year after diagnosis. Other specialty care visits were similar across groups in this same year (Figure 1).

When comparing changes in visits from prediagnosis to 1 year after diagnosis, urology visits increased substantially, as did oncology visits, especially for the “high cancer” group. Although urology visits decreased 2 years after diagnosis for all three groups, oncology visits remained elevated for the “high cancer” group. There was little change across years for primary care and other specialty care visits for all three groups.

Health-Care Utilization for CRC Patients

For CRC patients, we included both oncology and gastroenterology/surgery (GI/surgery) visits separately to represent care that is more likely associated with initial cancer treatment. When comparing the three probability groups for CRC, oncology visits in the year after diagnosis were highest for the “high cancer” group (36.2 visits/1000 person-days) followed by the “equal” and “high other” groups (9.2 and 5.4, respectively). GI/surgery visits increased in the year after diagnosis with close to the same rate across the three probability groups (12.5, 11.6, and 9.5 visits/1000 person-days for the “high cancer,” “equal,” and “high other” groups, respectively). Primary care visits in the year after diagnosis were highest among the “high other” group (10.9 visits/1000 person-days) and were approximately the same in the “equal” and “high cancer” groups. Other specialty visits were close to equal across the three probability groups. We did not see large change in number of visits from prediagnosis to the first and second year after diagnosis for primary or other specialty care visits. There was a large increase in oncology and GI/surgery visits from prediagnosis to the first year after diagnosis and with a drop in visit numbers for the second year after diagnosis for all three groups.

Discussion

In a cohort of KPCO prostate cancer and CRC patients, we found an increase in the number of cancer-related specialty visits (oncology for both prostate cancer and CRC, urology for prostate cancer, and GI/surgery for CRC) in the year after diagnosis. The number of visits was highest for the groups with high probability of dying of either type of cancer. This increase in the number of oncology visits was sustained for the “high cancer” prostate cancer group through the second year after diagnosis but decreased for the other subgroups. We found that during the year after cancer diagnosis, primary care and other specialty care utilization appeared to vary little across most subgroups. Likewise, visits to primary and other specialty care remained stable in the 1 and 2 years following cancer diagnosis. Thus, it did not appear that either a new cancer diagnosis or the prognosis of that cancer affected the extent of contact between these patients and their clinicians.

Increase in the utilization of oncology-related health-care services after cancer diagnosis in our cohort is not surprising and has been documented by previous studies (12). However, prior studies have reported mixed findings regarding the utilization of primary care utilization following cancer diagnosis (6,7,12–15). Our study showed that primary care and noncancer care utilization did not differ following diagnosis of cancer and might be explained by coordination of primary and specialty care services within an integrated health-care system or ongoing care for noncancer comorbidities (16). Given the intensely busy and stressful period following cancer diagnosis, effective coordination of care focusing on continued management of chronic medical conditions and prevention of unnecessary visits may be optimal in achieving patient-centered care (17).

A number of previous studies have suggested that patients with multiple comorbid conditions tend to use more medical care (18–21). In our multimorbid sample, we found that the severity of the patient’s cancer (higher likelihood of dying of cancer 5 years within diagnosis) appeared to be accompanied by an increase in primary care visit numbers for CRC, whereas the likelihood of dying of other comorbidities did not appear to affect primary care and other specialty care utilization.

The decrease in health-care utilization in oncology and primary care in the years following cancer diagnosis has been previously reported by Keating et al. (22). Keating et al. documented a decrease in the role of care provided by both primary care physicians and oncology specialists over a period of 3 years following diagnosis. Our study suggests that although oncology utilization decreases, primary care visits remain stable. This is likely emphasizing the potential benefits of an integrated health-care system whereby patients are not lost in transition following completion of therapy (23).

As a new approach to looking at utilization patterns among cancer patients, our analysis stratified patients by their type of primary morbidity burden. We found that although specialty care peaked after diagnosis for those with a high cancer morbidity burden, it appeared to increase less for those with a high other morbidity burden—perhaps reflecting care choices and priorities. Furthermore, primary care use appears to be about the same before and after cancer diagnosis within morbidity burden strata suggesting that in a population with a moderate number of comorbidities, primary care use may not drop off in the face of the potential competing demand of a cancer diagnosis.

This study has several limitations. The purpose of our analysis was primarily exploratory and descriptive and we did not use statistical tests to compare subgroups of patients or time points. Although this approach limits the conclusions we can draw from this analysis, it still provides valuable data to inform future hypothesis testing studies. Furthermore, our population derived from an integrated delivery system and the relevance of our findings, particularly with respect to service utilization, may not translate to other groups of patients. We also made an assumption that other specialty care visits are less likely to reflect initial cancer treatment-related visits. However, we anticipate that the misclassification resulting from this assumption would not affect our main conclusions. Future studies should compare trends statistically and explore the impact of mortality risk on health-care utilization patterns in a nonintegrated care context.

Conclusions

Feuer et al. (9) found that nomograms developed for prostate cancer and CRC using a population-based sample remained valid when applied to a clinic-based cohort patient. When using these nomograms to classify cancer patients based on their predicted probability of dying of cancer and other conditions, we found that although a new diagnosis of cancer increased utilization of cancer-related services for an extended time period, the timing of cancer diagnosis did not appear to affect other types of utilization, which may in part reflect total morbidity burden. Together, the findings from Feuer et al. (9) and the analysis presented in this article provide information on the validity and usefulness of cancer nomograms developed using a population-based sample in a managed care setting. Future research should assess comparative causes and effects of differential use of health-care services after cancer diagnosis. It should also explore how prognosis may be related to treatment burden for cancer and other comorbidities.

Funding

This work was supported by the Agency for Healthcare Research and Quality (R21 HS19520-01 to BAR, JLE, JFS, and EB) and the National Institutes of Health (P20 CA137219 to BAR and HHSN261201200414P for BFH).