This study examined relationships among cancer, comorbidity, and physical and mental health in the SEER-MHOS data linkage project. Results indicated that across most cancer types, after adjusting for demographic differences, individuals with a cancer history have a small, but significantly higher, prevalence of most of the comorbid medical conditions measured on the MHOS. After also accounting for these medical comorbidities and the time since cancer diagnosis, results showed that cancer patients (other than those with melanoma) had significantly worse physical health compared with patients without cancer. Non-small cell lung, NHL, breast, colorectal, and bladder cancer patients also reported worse mental health than did patients without cancer. On further investigation of the four most prevalent cancers in the U.S. (prostate, breast, colorectal, and lung cancer), we saw that negative associations with physical and mental health were most pronounced in those with two or more comorbidities, and in those diagnosed with cancer within the past year. Thus, our main hypotheses were supported. These results highlight the importance of examining associations between cancer and HRQOL by specific cancer types, and accounting for the recency of diagnosis and presence of other medical comorbidities. The SEER-MHOS data linkage provided a unique opportunity to investigate these relationships in a large sample of older adults.
Overall, cancer patients had a high prevalence of at least one comorbid condition (ranging from 84 to 88 percent) as did individuals without cancer (83 percent). These data are generally consistent with previous studies (Koroukian, Murray, and Madigan, 2006
; Ogle et al., 2000
). When compared with patients without cancer, the magnitude of the differences in our study were modest, with cancer patients in this sample having only 1 to 3 percent higher prevalence estimates for other comorbid conditions. This is not surprising, as many cancers and other chronic health conditions have similar risk factors such as smoking, obesity, physical inactivity, and poor diet. Previous estimates using national data have shown that overall, a higher percentage of cancer patients report at least one comorbid condition compared with patients without cancer (Hewitt, Rowland, and Yancik, 2003
). However, prevalence estimates were not as disparate among individuals age 65 or over. In our study, when the number of comorbid conditions was examined by cancer type, it became clear that among patients with certain cancers, particularly melanoma and NHL, the adjusted number of reported comorbid conditions was no higher than for patients without cancer, and that prostate cancer patients had only a marginally higher mean number of comorbid conditions. Conversely, among individuals with kidney, lung, endometrial, bladder, colorectal, and breast cancer, the number of comorbid conditions was significantly higher. The percentage of patients with these cancer diagnoses reporting one or more comorbid condition was approximately 3 to 6 percent greater than was the case for patients without cancer. These data suggest that it is important to separate out types of cancer when investigating the role of other comorbid medical conditions in HRQOL.
Examination of physical and mental health indicators revealed that across most cancer types, cancer patients had statistically significantly lower scores than did patients without cancer after accounting for other medical comorbidities, the time since cancer diagnosis, and other relevant characteristics such as age. Results from this study also provided some information about the relative impact of chronic diseases, including cancer, on HRQOL. In general, results showed that most non-cancer comorbidities had stronger associations with both PCS and MCS than cancers (i.e., they generally had larger coefficients) and were more efficient predictors, evidenced by the smaller standard errors. Data indicated that arthritis and lung disease had the greatest impact on physical health, whereas gastrointestinal disorders and strokes had the greatest negative impact on mental health. These findings are important, as they indicated that cancer patients may have many competing comorbidities that are relevant to HRQOL, and in some cases have a stronger impact than the cancer itself.
Closer examination of HRQOL in cancer versus non-cancer patients indicated that compared with patients without cancer, lung and NHL patients reported the worst physical and mental health. However, in samples this large, it is also important to examine the magnitude of the differences as well as their statistical significance. Different criteria have been used in the literature to indicate the minimally important difference (MID) necessary to signify a meaningful or clinical effect (Guyatt, Walter, and Norman, 1987
). Cohen’s (1992)
criteria suggest that a small effect is indicated by a 0.20 SD and a 0.50 SD is a medium effect size. There is support in the literature that MIDs fall within this range (Hays, Farivar, and Liu, 2004
; Kosinski et al., 2000
). In this case, a 2 to 5 point difference, or greater on the PCS and MCS would be considered large enough to be important.
For both lung cancer and NHL, physical health scores were close to one-half of a SD (5 points) lower than were those of patients without cancer. These differences indicate a medium effect, and exceed estimates of MIDs for the SF®
-36 (Ware et al., 1993
; Hays and Morales, 2001
). So in addition to statistical significance, these results also suggest clinically meaningful differences in some cancers. The importance of identifying this difference for is because cancer care often fails to address patients’ HRQOL issues in cancer survivors (Institute of Medicine, 2008
). These findings suggest that health care providers, particularly those focused on caring for lung and NHL survivors, should provide appropriate support services to attenuate the adverse HRQOL consequences of cancer and its treatment. Additionally, these findings illustrate the need for policymakers and payers to make it possible for clinicians to provide these services by reimbursing them for this service. It is important for health care providers to consider the impact of cancer on HRQOL when caring for patients with these cancers. In addition, it is important for policymakers to begin to support the integration of HRQOL data in clinical care.
We were able to further explore associations between comorbidities and time since cancer diagnosis within the most prevalent cancers (prostate, breast, colorectal, and lung cancer patients). Results indicated that individuals with the greatest number of comorbid conditions, who were also diagnosed with cancer in the previous year, had the worst HRQOL across both physical and mental health. Those individuals recently diagnosed with cancer are likely to be in treatment or recovering from treatment and its associated acute side effects. It is possible that more attention needs to be paid to controlling these symptoms as they are known to adversely affect HRQOL (Hodgson and Given, 2004
; Kurtz et al., 1999
). Further, it appeared that having a greater number of comorbid conditions was a stronger indicator of lower PCS and MCS scores than was time since diagnosis. These findings have important implications, particularly for long-term cancer care. While those closer to diagnosis are typically managing side effects of cancer and treatment, and therefore report worse physical and mental health, these results suggest that multiple chronic conditions may have important effects in long-term survivors.
Although results showed fairly consistent patterns across cancer types, differences were most pronounced among lung cancer patients, where the difference in scores for those with the most comorbid conditions, diagnosed in the past year, to those with no comorbid conditions diagnosed more than 5 years ago, was 17 points on the PCS and 12 points on the MCS. However, lung cancer patients who survive for 5 years or who have no comorbid conditions are likely to be very different from most lung cancer patients. Lung cancer patients age 65 or over have a 5-year survival rate of 14 percent (Surveillance, Epidemiology, and End Results Program, 2007
) and the prevalence of comorbid conditions associated with smoking such as cardiovascular disease and COPD have been shown to be higher than the general population (Janssen-Heijnen et al., 1998
). While it appears that long-term survival is associated with better HRQOL in lung cancer patients, this is likely due to a healthy survival effect, as individuals with greater disease severity were not as likely to survive. However, in general, given lung cancer patients’ overall poor prognosis, it may be important for clinicians to consider appropriate palliative care to this group. For prostate cancer, where HRQOL was highest, the difference in scores for those with the most comorbid conditions who were diagnosed with cancer in the past year, to those with no comorbid conditions diagnosed more than 5 years ago, was still a wide margin at 10 points on the PCS and 4 points on the MCS. The additional disease burden in those with comorbid conditions has important implications for physical and psychological limitations in individuals with cancer. These results highlight the importance of assessing other chronic diseases when caring for cancer patients. Results also suggest that PCS and MCS scores are higher with increased time since diagnosis, although these results cannot conclusively determine longitudinal effects, given the cross-sectional nature of the data. These results underscore the need to specifically tease out the effects of comorbid conditions, time since diagnosis, and cancer-specific complications, when examining relationships between cancer and HRQOL in older adults.
Our results have a number of clinical implications. Among the various cancer types, kidney and lung patients reported the greatest number of comorbid conditions, and had major decrements in physical and mental health as compared with patients without cancer. These results may be due to several factors, including comorbid conditions, effectiveness of available treatment, complications of aggressive treatment, or the possibility of being diagnosed at a later stage (which is common among lung cancer patients). It also may be important to explore whether comorbidities are a consequence of cancer or whether pre-existing comorbidities interact with cancer treatment, or both. For example, a recent randomized clinical trial of androgen-suppressing prostate cancer therapy showed an interaction between comorbidity and survival outcome. Overall, survival with the therapy was improved 80 percent. However, reduced mortality was seen only in males who had little or no comorbidity when they began treatment (D’Amico et al., 2008
). These data highlight the importance of ensuring that patients’ HRQOL needs are being addressed appropriately. Given the complexities of these relationships, it may be important to include a geriatric consultation or have geriatricians in the health care team to help manage the health care needs of older cancer patients (Gianni et al., 2001
). This is especially important when designing supportive care interventions, particularly behavioral interventions such as those promoting physical activity. Clinicians need to be prepared to address the comorbidities that, combined with cancer, may significantly influence care and health outcomes.
The current effort extends previous research using MHOS data (Baker, Haffer, and Denniston, 2003
; Ellis et al., 2004
; Ko and Coons, 2005
), by examining associations between several types of cancer and physical and mental health. Data from Baker and colleagues (2003)
were based on the 1998 MHOS and data from Ko and Coons (2005)
were based on the 2001 MHOS. Both suggested that cancer (grouped as one category) does not have nearly as strong associations with HRQOL scores as do other comorbid conditions. However, our data suggest that there are strong differences by type of cancer. In particular, lung cancer and NHL were both among the top five medical conditions associated with worse physical health, and lung cancer also had a strong negative relationship with mental health. Conversely, melanoma had very little impact on either physical or mental health, which is not surprising given that melanoma treatment is often less aggressive, and the clinical course for early stage melanoma is more favorable than for other cancer types. Taken together, our results suggest that it is very important to examine the effects of cancer by particular cancer type.
Although these data have a number of strengths, they also have limitations. Due to the cross-sectional design of this analysis, we are unable to determine whether cancer patients are more likely to develop comorbid conditions after their diagnosis or whether they are pre-existing. Further, we cannot determine whether cancer patients are more likely to be diagnosed with comorbidities due to having more medical scrutiny or as a complication of cancer or its treatment. The MHOS does not provide information about the severity of comorbid conditions. It also only asks about 12 conditions and therefore does not capture such common conditions in the elderly as osteoporosis, benign prostatic hypertrophy, or dementia. Another limitation is that HRQOL was assessed using a generic measure, whereas a cancer-targeted instrument might have yielded information on symptoms particularly relevant for cancer patients. Further, we did not examine cancer treatment or staging information, and we did not have an objective measurement of comorbid conditions. However, data comparing MHOS survey items with medical records suggests that patients can provide reasonably good reports of their morbidity in survey questions (70 to 94 percent specificity and 65 to 85 percent sensitivity) (Miller et al., 2008
The current analysis did not use a matched-design (such as the methods described by Reeve et al., 2008
), but instead included relevant characteristics as covariates in the models. However, there is evidence to suggest that adjusting for differences compared with matched propensity score designs often lead to the same conclusions (Rubin, 1979
), particularly if groups are similarly distributed on the covariates (e.g. race, sex, age, etc.). Matched designs are more important when one group has a disproportionate representation of particular characteristics (e.g., one group is mostly Black, the other is mostly White; Rosenbaum and Rubin, 1985)
. In our sample, patients with and without cancer differed on a number of characteristics, but not dramatically. For example, cancer patients were slightly, but significantly older (approximately 1 year older on average) than patients without cancer. Finally, it is important to note that when examining time since cancer diagnosis as an important factor in predicting HRQOL, we are unable to tease apart a healthy survivor effect in which those who lived longer were healthier for a variety of unmeasured reasons. This is particularly true for lung cancer patients; 29 percent of lung cancer patients in this sample had survived for more than 5 years, while national survival rates are 14 percent. Thus, our findings may not generalize to those individuals with worse prognoses, although it is likely that longer survival would continue to be associated with better HRQOL. Future analyses should examine effects of cancer-specific variables, such as time since diagnosis as well as other medical comorbid conditions on survival.
Overall, this analysis adds to the existing cancer and HRQOL literature by comparing patients with several different cancer types (including many types that are less prevalent in the U.S. population) to patients without cancer. It was further able to demonstrate the importance of examining both the number of comorbidities and the time since a cancer diagnosis. These data clearly support the need for clinicians to assess other medical comorbidities when treating cancer patients and developing supportive care interventions. Future research should extend these findings to more specific measures of HRQOL and examine other cancer-specific variables such as type, stage, and treatment.