Fax: (517) 353-9553
Depression and functional status as predictors of death among cancer patients
Article first published online: 14 MAY 2002
Copyright © 2002 American Cancer Society
Volume 94, Issue 10, pages 2719–2727, 15 May 2002
How to Cite
Stommel, M., Given, B. A. and Given, C. W. (2002), Depression and functional status as predictors of death among cancer patients. Cancer, 94: 2719–2727. doi: 10.1002/cncr.10533
- Issue published online: 14 MAY 2002
- Article first published online: 14 MAY 2002
- Manuscript Accepted: 14 DEC 2001
- Manuscript Revised: 22 OCT 2001
- Manuscript Received: 7 MAY 2001
- National Institute of Nursing Research
- National Cancer Institute
- physical functioning;
- cancer survival;
- duration of limitations
The current study examined the extent to which depression and functional limitations contribute to the mortality of newly diagnosed cancer patients. The analysis focused on differences in survival times among cancer patients with new experiences of depressive symptoms and functional limitations and patients with a history of such limitations.
Data for the current analysis came from two panel studies conducted in Michigan between 1993 and 1997, including 871 adult (≥ 21 years of age) breast, colon, lung, and prostate carcinoma patients. Information came from four separate sources: the intake patient interview, a self-administered questionnaire, medical record audits, and the Death Certificate Registry of Michigan's Department of Community Health. With time to death as the primary outcome (followup of 571 days), data were analyzed using Kaplan-Meier product limit estimates and the Cox proportional hazard model.
Cancer patients who, after diagnosis, report only new depressive symptoms or functional limitations, have the same survival chances as those who report none. Cancer patients with either previous emotional problems or previous physical limitations face, within the first 19 months after diagnosis, a 2.6 times greater hazard of dying than patients without prior problems. Patients with both previous emotional problems and physical limitations before diagnosis have a 7.6 times greater hazard of dying within that time frame.
The current data show cancer patients with prior limitations and emotional problems have worse survival chances than would be expected on the basis of their cancer diagnosis alone. While depressive symptoms and functional limitations are common short-run responses to a cancer diagnosis and initial treatment, patients with no prior history of such problems appear to be more resilient. Cancer 2002;94:2719–27. © 2002 American Cancer Society.
Recent literature on the value of quality of life measures as prognostic indicators of survival among cancer patients seems to indicate that such measures could serve as successful predictors of survival. 1–4 While quality of life measures invariably include mental health and physical functioning measures as subscales, they are designed to measure ultimate outcomes as evaluated by the respondents. Measures of depression and physical or cognitive functioning represent both outcomes in their own right and intermediate variables that may affect and modify perceptions of quality of life. Since both depression and physical functioning are, in principle, amenable to interventions, their relationship to survival takes on added clinical significance. There are several studies in which researchers have employed separate measures of self-rated physical functioning and have shown such ratings to predict survival independently of medical diagnostic information, 5–7 although such findings are not uniform. 8 The impact of depression or depressed mood on survival also remains unclear. Several studies have shown depression symptoms to be predictors of all-cause mortality, 9–11 yet others have not been able to confirm such a pattern. 12, 13
For a better understanding of these contradictory findings, we need to focus more on the causal status of measures of physical functioning and depressive symptoms. One way to view them is simply as indicators of the general physical and psychological functioning of a person at a particular time. As such, they are not independent causes of mortality but reflect the underlying health status of persons, including the consequences of specific diseases from which they may be suffering. Viewed in this way, measures of physical and psychological functioning can be used to summarize health status and predict and elucidate the trajectory towards death. 14
Another view of such quality of life measures is that they represent independent contributors to the risk of death. For instance, in a sample of Mexican-Americans diagnosed with cancer, Black and Markides 15 showed depression to be an additional risk factor that reduces the survival chances of cancer patients. In this context, depression is thought of as more than a summary indicator of health status. Instead, it may be seen as a case in its own right, since depression as a psychological syndrome often involves the loss of the will to live. 16, 17 This line of reasoning appears less plausible in the case of self-rated physical functioning measures. Physical limitations may not so much cause a person to die earlier, but a person's acknowledgement of her/his own frailty may indicate the presence of as yet undiagnosed specific illnesses and co-morbid conditions. In this sense, measures of self-rated physical limitations seemingly make independent contributions to explaining the risk of mortality, 6 even though they may well be proxies of undetected health conditions. Supporting, though not conclusive, evidence for the independent contribution thesis may be seen in the numerous studies that have shown physical functioning and depression symptoms as independent predictors of mortality, even after controlling for medical diagnosis. 6, 18, 19 However, many of the studies focusing on mental health predictors of survival among cancer patients are based on small clinical samples or control for very few potential confounders. 20–22
That still leaves the question open as to how physical functioning and depression on one side and specific illnesses, such has cardiovascular diseases or various cancers, on the other side are related to each other. For instance, depression appears to be a risk factor for overall mortality, 20, 21 but it is not clear what the significance of such a finding is. One possibility is that depressive symptoms are simply a response to many diseases, especially if these diseases are life threatening and have a poor survival prognosis. However, while the onset of depression may be caused by the disease, depression may assume a mediating role, resulting in a different trajectory of the disease among depressed patients. 22 Another possibility is that a person may already be depressed (or have physical limitations) prior to the onset of a specific new disease, and that may make recovery prospects worse. In that case, even though there is no causal link between depression and the specific disease in question, 22 the presence of depression modulates the course of the disease. Finally, depression in particular may be a precursor of and actually have contributed to the contraction of a specific disease. For instance, there is some evidence that a depressive disposition accelerates the onset of cardiovascular diseases, 23, 24 although it does not seem likely that it is a risk factor for the occurrence of cancer itself. 24, 25
Given these possibilities, results from cohort studies that employ baseline quality of life measures to predict survival among patients with specific medical diagnoses are usually difficult to interpret. Most studies (with exceptions, such as Penninx et al. 22) do not make a distinction between depression and functional limitations as responses to the initial diagnosis (and treatment) or as long-standing co-morbid conditions. Using recall measures about patient states prior to the diagnosis, the current analysis examines if it makes any difference for survival whether depression and functional status limitations are new experiences among newly diagnosed cancer patients or whether they are long-standing phenomena. The analysis takes into account clinical markers, such as staging of the cancer and primary diagnosis, as well as other known predictors of mortality, such as age, gender, socio-economic status, and a variety of chronic physical co-morbid conditions.
The data for the current analysis came from two panel studies of cancer patients and their caregivers conducted in Michigan between 1993 and 1997. The studies employed almost identical data gathering instruments and study designs (4 panel interviews spaced over 12 months), but differed somewhat in their eligibility criteria. The community-based study included newly diagnosed cancer patients who were 65 years old or older and were diagnosed with one of the four primary cancer diagnoses: breast, colon, lung, and prostate carcinomas. For the rural-partnership study, eligibility criteria covered any adult (≥ 21 years of age), newly diagnosed person with any site of cancer residing in eight rural counties of southwestern Michigan. However, for the current analysis, only data from patients with the four major diagnoses were used.
The cancer patients in the studies were recruited from a wide range of community hospitals as well as medical oncology and radiation oncology clinics (n = 67) located in lower Michigan outside the Detroit metropolitan area. Recruitment for the subjects used in the current analysis occurred between February 1994 and March 1997 and included all eligible cancer patients who were newly diagnosed and were treated at the 67 sites. While the study sample is not a probability sample of newly diagnosed cancer patients in lower Michigan, the subjects came from every corner of Michigan's lower peninsula. They were resident in 343 different zip code areas of 289 different Michigan townships, which are located in 66 of the 68 counties of the lower peninsula. After written patient/family consent was obtained, families were contacted for their first telephone interview, usually six to eight weeks after the initial cancer diagnosis. Study procedures fulfilled all requirements of the University Committee on Research Involving Human Subjects as well as of the Institutional Review Boards at participating institutions.
Consent to participate was provided by 1200 patients or caregivers in the community-based study and 210 patients or caregivers in the rural-partnership study (the number 210 refers only to those patients with the four major cancer diagnoses). Data for the current analysis came from four separate sources: the intake (wave 1) interview of patients, a self-administered mailed patient questionnaire, medical record audits, and the Death Certificate Registry of the Michigan Department of Community Health. In part because the study design allowed for temporary case participation of caregivers without their patients, in part because many patients did not return the mailed questionnaires, information from all four data sources was available on 940 out of 1410 cases (66.7%). Among these 940 cases, 871 (92.7%) had complete information on all variables used in the analysis.
All information pertaining to socio-demographic characteristics of the patients and families involved came from the patient (or caregiver) intake interviews.
The ultimate outcome measure in the current analysis is the survival status of the participating study subjects for up to 19 months after consent and initial enrollment into the study. Death or survival was, at first, determined through re-contacting families. In addition, we confirmed all death information and established the death of a few cases lost to followup using information from the Death Certificate Registry of the Section on Vital Records and Health Data in the Michigan Department of Community Health. These data were obtained seven months after the end of the panel study, thus accounting for the 19 month period of observation from the time of consent.
Health status variables
Primary cancer site. Information on the primary diagnosis or primary site of cancer was obtained from both the interviews and medical record audits. With few exceptions, information was consistent, and six discrepancies were resolved.
Staging. Information on the staging of the cancer was obtained from medical records alone. All cases were classified on the basis of the TNM system. Staging categories were collapsed into a simple dichotomy of late-stage (TNM Stages III and IV and metastatic cancer) and early stage (TNM Stages in situ, I, and II).
Chronic co-morbid conditions. Information on co-morbid conditions was obtained from the patient interviews. Patient reports have been shown to be a reliable source of such information. 26 Among the list of co-morbid conditions in the interview, we selected six physical conditions with established effects on longevity: heart disease, high blood pressure, diabetes, chronic lung disease, stroke, and fractured hip. The measure employed consisted of a simple count of the reported chronic co-morbid conditions.
Physical functioning. Physical functioning was measured using a sub-scale from the Medical Outcomes Study 36-Item Short Form Health Survey (SF-36). The Physical Functioning Sub-scale (PFS) of the SF-36 consists of 10 items (Cronbach's Alpha = 0.90), including measures of the degree of limitation in activities such as lifting or carrying groceries, bending, kneeling or stooping, etc. The scores for the sub-scale were standardized in the usual manner on a scale of 0–100. 27 For ease of graphic presentation, the scale was dichotomized at the cut-off point of 95 (95+: no physical limitations, < 95: some or substantial physical limitations). The same procedure was followed for the recall measure of the PFS. For the recall measure, subjects were asked to rate their functional status three months prior to the interview, i.e., before the diagnosis. While concurrent ratings of functional status prior to diagnosis (and identification of the patient) were impossible to obtain, it may actually be an advantage to rely on a recall measure of prior functional status. In that way, both before and after diagnosis measures are rated using the same reference point. This makes the occurrence of a response shift less likely. 28, 29 It seems unlikely to us that patients who experience new functional limitations project them onto the pre-diagnosis phase. If so, this would lead to a reduction in the count of patients with new limitations and false positives among those with already existing limitations. Combining prior and current functional status measures resulted in three categories: 1) no current functional limitations, 2) new limitations (a score of 95+ three months prior to intake interview but of < 95 at the time of the interview), 3) already existing limitations (already a score of < 95 for the recall measure). The theoretical possibility of improvement in functional status between pre- and postdiagnostic phases did not occur in this study sample.
Depression. Depressive symptomatology was measured using the Center for Epidemiological Studies Depression Scale (CES-D). This 20-item scale (Cronbach's Alpha = 0.89) was scored in the usual manner, with a potential range for the summated scale score from 0–60. For the survival analysis, we chose the customary cut-off point of 16 or higher to divide the sample into those with likely presence of depressive symptoms and those without such symptoms. 30 In addition, patients were asked if they had ever seen a health professional for emotional, nervous, or psychiatric problems. Those with current depression symptoms who answered no were classified as having new depressive symptoms; those with current depression symptoms who answered yes were classified as having continuing/existing depressive symptoms. All other patients were classified as having no current depression symptoms.
|Patient marital status|
|Patient living arrangement|
|Two- or multi-person household||855||78.3%|
|Patient health insurance|
|Medicare and medigap||700||80.4%|
|Patient age in years||70.1||71.0||8.4||28–98|
|Patient education in years||12.7||12.0||3.1||4–20|
|Patient household income||$28,851||$22,500||$18,980||$4–100,000|
|Survival status within 19 months of consent|
|Primary carcinoma diagnosis|
|Stage at diagnosis:|
|Chronic co-morbid conditions:|
|Two or more||404||46.4%|
|Presence of physical limitations at wave I interview (based on SF-36 PF cut-off point of 95+)|
|Presence of depressive symptoms at wave I interview (based on CES-D cut-off point of 16+)|
|No depressive symptoms||591||67.9%|
|New depressive symptoms||198||22.7%|
|Existing emotional problems||82||9.4%|
|SF-36 physical functioning||64.6||70.0||28.7||0–100|
|Patient CES-D score||11.5||10.0||8.3||0–57|
|Days until death (n = 151)||275.7||253.0||153.8||25–570|
The following analyses of predictors of survival among newly diagnosed cancer patients involved right censored data: information about subjects' death was followed for up to 19 months after intake into the studies. With time to death as the primary outcome and a maximum followup of 571 days, the results are presented in two stages: univariate analyses using Kaplan-Meier product limit estimates and multivariate analyses involving the Cox proportional hazards model. 31 All statistical analyses were performed using various survival time commands of STATA 7.0. 32
Figure 1 shows a comparison of the non-parametric (Kaplan-Meier) survival curves among three categories of cancer patients. 1) Cancer patients with no physical limitations (scoring at or above 95 on the standardized PFS at the time of the intake and on the three month recall measure) are compared to 2) cancer patients with new physical limitations (scoring below 95 on the current but not the recall measure of the PFS) and 3) cancer patients with continuing physical limitations (scoring below 95 on both the current and recall measures of the PFS). In this comparison, the survival rates of long-term functionally limited cancer patients are consistently lower than the rates of patients with new physical limitations which, in turn, are lower than the survival rates of cancer patients with no reported physical limitations. The resulting final survival proportions are 74.8%, 89.2% and 95.7% respectively. The log rank and Wilcoxon-Breslow tests indicated that the survival experience differed among these three groups of cancer patients. For a similar comparison of Kaplan-Meier survival curves, the cancer patients were further divided into three categories based on their depression status shortly after diagnosis: 1) subjects with new depression symptoms, scoring 16 or higher on the CES-D at the time of the intake interview but having no history of emotional problems, 2) subjects with current symptoms and a history of emotional problems, and 3) subjects with no current depression symptoms. The mean (restricted) survival time of cancer patients who reported new depression symptoms and had no history of emotional problems appeared to be the shortest (mean of 468 days) among the three groups. By contrast, patients who reported no depression symptoms at the intake interview had the best survival chances (mean of 538 days), while patients with a history of emotional problems and current depression symptoms were somewhat in-between (mean of 515 days). Accordingly, the proportions of survivors at the end of the observation period (after 19 months) differed substantially for the three groups: 70.7% among the newly depressed, 81.8% among the patients with existing emotional problems, and 86.8% among the not depressed patients. The log rank (31.29, df: 2, P < .001) and Wilcoxon-Breslow (33.51, df: 2, P < 001) tests again indicated that the survival experiences differed among these three groups of cancer patients.
The analysis presented in Table 3 sheds light on the efficacy of depression symptoms and functional status in predicting patient survival in a multivariate context. The Cox proportional hazards model was fitted to the data, with six socio-demographic control variables (gender, age, education, race, living arrangement, and patient household income) and three medical/diagnostic control variables (primary cancer site, staging at diagnosis, and the number of chronic, physical co-morbid conditions). The data met the proportionality assumptions of the Cox model: all three regression models of Schoenfeld residuals on functions of time, using Kaplan-Meier, rank, and logscaling, produced non-significant slopes. 33 Table 3 only shows the hazard ratios (HR), standardized regression coefficients (Z), the associated significance levels, and the 95% confidence intervals for the hazard ratios. The model only encompasses main effects, since all two way interactions, except the one involving depression and physical functioning, were statistically insignificant. (The interaction effect will be examined below.) As the Z statistics show, staging of the cancer at diagnosis and primary cancer site have the largest impact on the hazard of dying. The estimated hazard of dying is almost five times (HR = 4.92) larger among late stage than among early stage cancer patients. Similarly, compared to lung carcinoma patients, breast carcinoma patients have an eight times lower hazard of dying (1/.12 = 8.33), while prostate and colon patients' hazards are six times (1/.16 = 6.25) and three times (1/.32 = 3.13) lower. In contrast, existing chronic co-morbid conditions like heart disease or diabetes do not appear to have an additional effect. (Models that included dichotomous indicators of each co-morbid condition separately did not show any effects either. Since the absence of co-morbid conditions is moderately correlated with the absence of physical limitations, some of the effect of co-morbid conditions may have been absorbed by the self-rated physical functioning measure.) Among the socio-demographic predictors, none seem to predict survival, not even patient age. Part of this may be due to the restricted range of some of these variables (over 90% of the subjects were 65+ years old and over 90% were white), but other variables like patient gender, educational achievement, and household income showed robust heterogeneity.
|Predictors||HR||Z||P of Z||95% CI of HR|
|Patient gender (1 = female, 0 = male)||1.00||0.02||0.984||0.69–1.46|
|Patient age (in years)||1.00||0.32||0.752||0.98–1.03|
|Patient education (in years)||1.05||1.34||0.179||0.98–1.12|
|Patient race (1 = minority, 0 = white)||0.99||−0.02||0.984||0.47–2.07|
|Living arrangement (1 = alone, 0 = with others)||0.81||−0.98||0.329||0.53–1.24|
|Patient household income (in $10,000)||0.91||−1.55||0.122||0.79–1.03|
|Chronic co-morbid conditions (none)a|
|1 co-morbid condition||0.91||−0.37||0.712||0.54–1.51|
|2+ co-morbid conditions||0.92||−0.36||0.719||0.57–1.48|
|Primary carcinoma diagnosis (lung)a|
|Staging at diagnosis (1 = late, 0 = early)||4.92||7.46||0.000||3.23–7.47|
|Physical functioning (not limited)a|
|Depression symptoms (no symptoms)a|
|New depressive symptoms||1.66||2.79||0.005||1.16–2.37|
|Prior emotional problems||2.04||2.41||0.016||1.14–3.65|
As far as physical functioning and depression symptoms were concerned, they did appear to predict survival among these cancer patients, even after accounting for the strong effects of diagnosis and staging. Compared to patients who did not report any physical limitations at intake, those with long-standing existing physical limitations had an almost three time greater hazard of dying (HR = 2.91), while those with new physical limitations appeared not to differ in their survival chances from those without physical limitations at intake. Among patients who reported depression symptoms at intake, both those with new depression symptoms and those with prior emotional problems had worse survival chances than patients who did not report depressive symptoms. However, beyond that there did not seem to be any difference in survival impact, whether or not the depression symptoms were new or existing (this contrast is not significant).
It was mentioned earlier that the interaction effect involving both physical functioning and depression symptoms was statistically significant. However, with two three-category variables, eight parameters are needed (3 × 3 − 1) to represent all interaction effects. Instead, we opted for recoding of the combined variables into four categories: 1) cancer patients with no physical limitations and no depression symptoms (n = 159; this is the reference category); 2) cancer patients who reported new physical limitations or new depression symptoms at the intake interview (n = 194); 3) cancer patients who reported either prior physical limitations or prior emotional problems (n = 74); 4) cancer patients who already had physical limitations and emotional problems before the cancer diagnosis (n = 44). A second Cox proportional hazards model was run, employing the same variables as in Table 2, except that the main effect variables for depression and physical functioning were replaced by the combined four-category interaction variable. Again, tests showed the proportionality assumption was met (Kaplan-Meier: P > .710, Rrank: P > .621, log: P > .510), and the overall model provided a good fit (chi square: 307.94, df: 15, P < .001). Not surprisingly, the hazard ratios for all the control variables were very similar to the previous hazard ratios. The interaction variable, combining physical limitation and depression symptom categories, showed a clear pattern: 1) cancer patients, who only reported new symptoms or limitations had the same survival chances as those who reported no symptoms (HR: 1.37, P < .447, 95% confidence interval [CI]: .60-3.16); 2) cancer patients who already had either previous emotional problems or previous physical limitations faced a 2.6 times greater hazard of dying than patients who reported no problems (HR: 2.61, P < .010, 95% CI: 1.25–5.45); and 3) cancer patients who already had both previous emotional problems and physical limitations before their diagnosis had a 7.6 times greater hazard of dying (HR: 7.61, P < .001, 95%CI: 2.89–20.03). The presence of both long-standing physical limitations and emotional problems had an exacerbating effect on mortality: the two interact to shorten survival beyond what would be expected from a mere multiplicative effect (7.61 > 2.91 × 2.04 = 5.94).
The analysis presented here adds to a growing body of evidence that, similar to quality of life measures, measures of physical functioning and depression symptoms reported early after cancer diagnosis appear to predict the survival trajectory of cancer patients. 6, 7, 15, 17, 34 The predictive value of self-rated functional limitations and depression symptoms on survival remained substantial, even after accounting for medical prognosis related to the specific cancer site and its staging as well as the presence of major chronic diseases. However, the use of recall measures in the current analysis allowed us to distinguish between patients with depression symptoms and physical limitations prior to the cancer diagnosis and those who experienced these problems in response to, or at least concomitant with, their cancer diagnoses. This distinction proved important: cancer patients who reported new depression symptoms and physical limitations at the intake interview (on average six weeks after diagnosis) possibly suffered from temporary problems without long-term consequences for their survival chances. Limitations in physical functioning at this early stage in the treatment trajectory are often the result of the lingering effects of surgery. 35, 36 Depression symptoms may also be short-run responses to the cancer diagnosis, just as is the case among many patients with cardiovascular diseases. 22 Thus, many cancer patients may experience an emotional shock shortly after diagnosis and initial treatment, but those not depressed before the diagnosis appear to be resilient. (For the 117 cancer patients with new depression symptoms and for the 31 patients with existing emotional problems, we had CES-D scores from a second survey three months later. The respective mean CES-D scores were as follows: new depression symptoms group: 21.3 [intake], 16.1 [wave 2]; existing –emotional problems group: 23.8 [intake], 21.2 [wave 2]. These data clearly suggest greater resilience among the first group.)
The current data show cancer patients with prior limitations and emotional problems have worse survival chances than would be expected on the basis of their cancer diagnosis alone. These findings differ from those of the only other study we could find that distinguished between emerging depressive symptoms and persistent depressive symptoms, albeit involving cardiovascular patients. 22 In that study, newly depressed patients were less likely to survive than chronically depressed patients. To what degree these findings are disease specific is not known. But it is not unreasonable to assume that, among already depressed subjects who also experience physical limitations, a cancer diagnosis and subsequent treatment may well lead to a loss of hope and a pessimistic outlook about their futures. 17
Despite their predictive power, it remains unclear to what extent the capacity to predict survival on the basis of measures of functional limitations and depressive symptoms after diagnosis is rooted in a cause-effect relationship. On the other hand, the fact that cancer patients who already experienced depressive symptoms and functional limitations prior to their cancer diagnosis had substantially worse survival prospects suggests the potential benefits of psycho-social interventions. For instance, if prior depression affects the disease trajectory for the worse, effective counseling interventions could indirectly result in substantial survival benefits. Only intervention studies that actually reduce depression and improve functioning may provide an answer.
- 14Psychological predictors of mortality in old age. J Gerontol. 1999; 54B: P44–P54., .
- 27SF-36 Health Survey manual and interpretation guide. Boston: The Health Institute, New England Medical Center, 1993., , , .
- 30Center for Epidemiologic Studies Depression Scale. In: KayserJD, SweetlandRC, editors. Test critiques. vol. 2. Kansas City: Test Corporation of America, 1986: 144–160., .
- 31Statistical analysis of epidemiologic data. 2nd edition. New York: Oxford University Press, 1996..
- 32StataCorp. Stata Statistical Software release 7.0. College Station, TX: Stata Corporation, 2001.