`a Predictor of Breast Cancer
`Survival
`
`Sandra E. Sephton, Robert M.
`Sapolsky, Helena C. Kraemer, David
`Spiegel
`
`Background: Abnormal circadian
`rhythms have been observed in patients
`with cancer, but the prognostic value of
`such alterations has not been con-
`firmed. We examined the association
`between diurnal variation of salivary
`cortisol
`in patients with metastatic
`breast cancer and subsequent survival.
`We explored relationships between cor-
`tisol rhythms, circulating natural killer
`(NK) cell counts and activity, prognos-
`tic indicators, medical treatment, and
`psychosocial variables. Methods: Sali-
`vary cortisol levels of 104 patients with
`metastatic breast cancer were assessed
`at study entry at 0800, 1200, 1700, and
`2100 hours on each of 3 consecutive
`days, and the slope of diurnal cortisol
`variation was calculated using a regres-
`sion of log-transformed cortisol con-
`centrations on sample collection time.
`NK cell numbers were measured by
`flow cytometry, and NK cell activity
`was measured by the chromium release
`assay. The survival analysis was con-
`ducted by the Cox proportional haz-
`ards regression model with two-sided
`statistical testing. Results: Cortisol
`slope predicted subsequent survival up
`to 7 years later. Earlier mortality oc-
`curred among patients with relatively
`“flat” rhythms, indicating a lack of
`normal diurnal variation (Cox propor-
`tional hazards, P = .0036). Patients with
`chest metastases, as opposed to those
`with visceral or bone metastases, had
`more rhythmic cortisol profiles. Flat-
`tened profiles were linked with low
`counts and suppressed activity of NK
`cells. After adjustment for each of these
`and other factors, the cortisol slope re-
`mained a statistically significant, inde-
`pendent predictor of survival time. NK
`cell count emerged as a secondary pre-
`dictor of survival. Conclusions: Pa-
`tients with metastatic breast cancer
`whose diurnal cortisol rhythms were
`flattened or abnormal had earlier mor-
`tality. Suppression of NK cell count and
`NK function may be a mediator or a
`
`marker of more rapid disease progres-
`sion. [J Natl Cancer Inst 2000;92:994–
`1000]
`
`Cancer poses numerous physical and
`emotional stresses. While disease and
`treatment exert a heavy physiological toll,
`accompanying anxiety about diagnosis
`and prognosis, taxing medical treatments,
`and disruption of social, vocational, and
`family functioning constitute a series of
`psychological stressors. Cancer patients
`repeatedly endure physical and emotional
`events that activate stress-response
`mechanisms, including the hypothalamic–
`pituitary–adrenal (HPA) axis. Such re-
`peated activation has been associated with
`HPA axis dysregulation and adverse
`health consequences (1).
`One sign of dysregulation in this en-
`docrine stress response system is altered
`circadian cortisol rhythms (2–4). In
`healthy individuals, cortisol levels are
`usually highest before awakening and de-
`crease during the day (5), but up to 70%
`of patients with advanced breast cancer
`show flattened circadian profiles, consis-
`tently high levels, or erratic fluctuations
`(6,7).
`Although the specific causes of circa-
`dian dysregulation in cancer are undeter-
`mined, cortisol dysregulation has been
`linked independently with the physical
`stress of cancer (8) and with psychologi-
`cal stressors (4). Among patients with
`breast and ovarian cancers, severe endo-
`crine disruption is seen in more advanced
`disease (e.g., patients with poor perfor-
`mance status and liver metastases have
`more markedly abnormal rhythms) (7,9).
`However, psychological distress also ac-
`companies aberration in cortisol levels
`among cancer patients (10). Physically
`healthy individuals with altered diurnal
`cortisol profiles have been characterized
`as chronically stressed (3,4), and aberrant
`rhythms are evident in subjects with de-
`
`Affiliations of authors: S. E. Sephton, Department
`of Psychiatry and Behavioral Sciences, University
`of Louisville School of Medicine, KY; R. M. Sapol-
`sky (Department of Biological Sciences), H. C.
`Kraemer, D. Spiegel (Department of Psychiatry and
`Behavioral Sciences), Stanford University, Palo
`Alto, CA.
`Correspondence to: Sandra E. Sephton, Ph.D.,
`Department of Psychiatry and Behavioral Sciences,
`University of Louisville School of Medicine, Lou-
`isville, KY 40292-0001 (e-mail: sephton@
`louisville.edu).
`See “Notes” following “References.”
`
`© Oxford University Press
`
`994 REPORTS
`
`JANSSEN EXHIBIT 2085
`Wockhardt v. Janssen IPR2016-01582
`
`
`
`Downloaded from
`
`http://jnci.oxfordjournals.org/
`
` by guest on October 3, 2016
`
`pression (11), unemployment (12), and
`post-traumatic stress disorder (2). Asso-
`ciations of cortisol profiles with psycho-
`logical stress suggest that rhythm dys-
`regulation may be a marker or mediator of
`psychosocial effects on cancer progression.
`Dysregulation of the cortisol response
`may compromise tumor resistance. Glu-
`cocorticoids have been implicated in tu-
`mor growth, both in animal and in in vitro
`studies (13,14). Cortisol may accelerate
`tumor growth via immunosuppressive ac-
`tions (15) or effects on metabolic pro-
`cesses (16). When cortisol profiles are af-
`fected, dysregulated patterns of immune
`activity and immune cell trafficking may
`also emerge (17,18). Indeed, breast and
`ovarian cancer patients with altered corti-
`sol rhythms show disruptions in patterns
`of circulating leukocytes, neutrophils,
`platelets, and serum proteins (9).
`The prognostic consequences of circa-
`dian disruption in cancer patients are as
`yet unknown (8,9). In this study, we in-
`vestigated HPA axis dysregulation as a
`predictor of metastatic breast cancer pro-
`gression. The primary hypothesis was that
`aberrant diurnal salivary cortisol rhythms
`would predict earlier mortality from
`metastatic breast cancer. Second, we con-
`ducted exploratory analyses to determine
`whether associations of cortisol rhythm
`disturbance with traditional prognostic in-
`dicators, medical treatments, psychoso-
`cial variables, and immune function ex-
`erted statistically significant effects in the
`cortisol rhythm–survival relationship.
`
`PATIENTS AND METHODS
`
`Study Population
`
`Women with metastatic breast carcinoma (n 4
`125) were recruited over a 5-year period from June
`1991 through January 1996 for a new randomized,
`prospective study of Supportive/Expressive group
`psychotherapy and cancer survival designed to rep-
`licate and extend our earlier finding of longer sur-
`vival with group support (19). Subjects had breast
`cancer with systemic metastasis and a Karnofsky
`rating of 70% or more, were fluent in English, and
`lived in the San Francisco Bay area. Subjects were
`excluded if they had one of the following: 1) active
`cancers within the past 10 years other than breast
`cancer, basal cell or squamous cell carcinomas of the
`skin, or in situ cancer of the cervix; 2) positive su-
`praclavicular lymph nodes as the only metastatic le-
`sion at the time of diagnosis; or 3) a concurrent
`medical condition likely to influence short-term sur-
`vival. The Stanford University Institutional Review
`Board approved the study procedures. Written in-
`formed consent was obtained from all participants.
`A total of 104 patients contributed data to the
`current analyses, while 21 were excluded because
`they were taking hydrocortisone-based medications
`
`(n 4 10), they did not provide saliva samples of
`adequate volume or number for analysis of diurnal
`cortisol profiles (n 4 8), or they reported feeling too
`ill to collect any saliva samples (n 4 3). Baseline
`endocrine data were collected after recruitment and
`before randomization (an average of 25 days passed
`from cortisol collection to randomization). Data
`concerning current medications, pain (Pain-Rating
`Scale) (20), depression (Center for Epidemiologic
`Studies Depression Scale) (21), and sleep quality
`(22) were also collected before randomization. Pa-
`tients had been diagnosed a mean (standard devia-
`tion [SD]) of 1.8 (3.1) years before the study entry
`(Table 1).
`Of the 104 patients in the current sample, 58 were
`randomly allocated to the psychotherapy group and
`46 were randomly allocated to the educational con-
`trol group. Baseline cortisol data were collected a
`mean (SD) time of 5.9 (1.4) years before the current
`analysis (range, 3.1–7.7 years). The mean survival
`time of the 71 deceased patients was 2.3 (1.5) years
`from the time of cortisol assessment (range, 0.3–7.1
`
`years), and the remaining 33 had been followed for
`a mean of 5.4 (1.6) years (range, 3.1–7.7 years). No
`patients were lost to follow-up.
`This study was designed to test two major hypoth-
`eses: 1) whether or not Supportive/Expressive group
`psychotherapy increased survival time, a replication
`of an earlier trial of the effect of this intervention on
`survival (19); and 2) whether endocrine dysfunction
`was associated with the rate of disease progression
`by examining the relationship between diurnal cor-
`tisol patterns and survival. In the current analysis,
`we tested only the second hypothesis rather than
`addressing treatment/control differences. At the time
`of analysis for this report, overall mortality was
`68%, a rate at which one could reasonably expect to
`see any relationship between a physiologic measure
`previously associated with advancing disease and
`survival time. However, 68% mortality is too early
`to test the group psychotherapy hypothesis. In the
`previous trial (19), the survival effect associated
`with group therapy did not begin to emerge until
`mortality had exceeded 50%. Were we to replicate
`
`Table 1. Demographic characteristics and description of 104 women with metastatic breast cancer
`enrolled in the study (data taken at study entry)
`
`Mean (standard deviation)
`
`53.2 (10.7)
`
`16.2 (2.6)
`
`4.0 (3.0)
`1.8 (3.1)
`
`Characteristic
`
`Age, y
`Ethnicity, %
`Asian, 5.8
`Black, 1.0
`Hispanic, —
`Native American, 1.0
`White, 90.4
`Other, 1.9
`Educational level, y
`Marital status, %
`Never married, 10.6
`Married, 52.9
`Separated, 1.9
`Divorced, 26.9
`Widowed, 6.7
`Other, 1.0
`Disease-free period, y
`Years from cancer recurrence to study entry
`Dominant cancer site at study entry, %
`Bone, 39.4
`Chest wall or regional lymph nodes, 30.8
`Viscera, 29.8
`Tumor estrogen receptor status, %
`Positive, 77.9
`Negative, 22.1
`Prior systemic treatment for metastatic disease, %
`Hormonal therapy, 77.9
`Chemotherapy, 46.2
`Currently on hormonal treatment at study entry, 69.2%
`Currently on megestrol at study entry, 12.5%
`Treatment within 2 mo of study entry, %
`Chemotherapy, 30.8
`Radiotherapy, 18.3
`Physicians’ Karnofsky performance rating, %
`70, 3.8
`75, 2.9
`80, 20.2
`85, 2.9
`90, 30.8
`95, 2.9
`100, 36.5
`
`Journal of the National Cancer Institute, Vol. 92, No. 12, June 21, 2000
`
`REPORTS 995
`
`
`
`Downloaded from
`
`http://jnci.oxfordjournals.org/
`
` by guest on October 3, 2016
`
`the previous findings, one would not expect the di-
`vergence to be statistically significant until well be-
`yond 50% mortality. Indeed, in the previous study,
`we waited until the mortality was greater than 95%
`before testing the treatment versus control group hy-
`pothesis (19). On the basis of our prior data, we are
`not yet able to conduct a valid test of the effect of
`Supportive/Expressive psychotherapy on cancer sur-
`vival. In fact, recruitment of the current sample
`ended only in 1995, and many of the survivors are
`relatively healthy at this point. A definitive result of
`the intervention will be published when the sample
`reaches the predetermined threshold of 90% overall
`mortality.
`
`Collection of Physiologic Data
`
`The measurement of cortisol in saliva has been
`confirmed to reliably reflect free cortisol levels in
`blood (23). Patients collected saliva at home with
`“Salivette” (Sarstedt, Inc., Newton, NC) devices,
`consisting of a cotton swab fitted into a plastic
`holder resting inside a centrifuge tube. Sampling
`was at 0800, 1200, 1700, and 2100 hours on 3 con-
`secutive days. Medications, sleep, diet, exercise, and
`stressors during the sampling period were assessed
`by questionnaire. Samples were centrifuged for 4
`minutes at 2500 rpm, and aliquots were frozen at
`−70 °C. From 95 patients who consented, blood
`samples for lymphocyte counts were drawn on 2
`days approximately 1 week apart between 0700 and
`1000 hours. Among healthy subjects, natural killer
`(NK) cell counts tend to peak during these hours.
`
`Laboratory Methods
`
`Cortisol levels were assessed by 125I radioimmu-
`noassay. Intra-assay coefficients of variation on
`three different saliva pools averaged 5.3%, and the
`inter-assay coefficient was 12%. Assay sensitivity
`was 0.008 mg/dL. Percents and absolute numbers of
`NK cells were measured by flow cytometry by use
`of monoclonal antibodies to identify cells that were
`positive for CD56 (NKH-1 clone) and negative for
`CD3 (CD3-T3 clone) cell surface antigens (both an-
`tibodies were received from Coulter Cytometry Inc.,
`Hialeah, FL). Cytolytic activity was assessed by the
`chromium-release method. Lytic units reflecting
`20% lysis (LU20) were calculated (24) and cor-
`rected for NK cell numbers. The mean absolute
`numbers of NK cells and the mean LU20 for the two
`specimens were used.
`
`Statistical Methods
`
`Because dysregulated cortisol rhythms have been
`characterized by flattened or aberrant daytime pro-
`files and healthy cortisol rhythms by high morning
`and lower evening levels, the slope of diurnal
`change in the cortisol level was calculated to esti-
`mate how each patient fit the normal (i.e., descend-
`ing) profile (25). Since the distribution of raw cor-
`tisol values is typically skewed and the normal
`diurnal profile may be approximated by an exponen-
`tial curve, raw values were log transformed. The
`regression of the 12 cortisol values on the hour of
`sample collection was calculated, with data pooled
`over the 3 days for each patient. Steeper slopes are
`represented by smaller b values for the slope of the
`regression, which indicate cortisol declining more
`rapidly. Flatter slopes (larger b values) indicate
`slower declines, abnormally timed peaks, or increas-
`
`ing levels during the day. The average area under the
`curve (AUC) was also calculated over 3 days of
`sample collection by trapezoidal estimation using
`log-transformed cortisol values.
`The primary hypothesis was tested by analyzing
`the relationship between the baseline diurnal cortisol
`slope and the subsequent survival time. A regression
`was conducted on survival time, measured from the
`date of saliva collection, by use of the Cox propor-
`tional hazards model in a two-tailed test. The sec-
`ondary hypotheses were exploratory in nature and
`were tested with conservative intent to rule out the
`possibility that any relationship between the cortisol
`slope and survival time was not explained by vari-
`ance due to some other factor related to the disease
`or treatment process. These hypotheses evaluated
`the possible contribution of traditional prognostic
`indicators, including age, site of recurrence, estro-
`gen receptor status, disease-free interval, time since
`diagnosis of cancer recurrence, and Physicians’
`Karnofsky Rating; medical treatments, including
`hormonal therapy (e.g., tamoxifen and megestrol),
`recent chemotherapy, and recent radiotherapy; psy-
`chosocial variables, including treatment versus con-
`trol group assignment, marital status, sleep quality,
`pain, and depression; and two immune measures,
`NK cell counts (absolute numbers per cubic milli-
`meter whole blood) and function (LU20). Spearman
`rank correlations with tie correction were used to
`test associations of the cortisol slope with each of
`these factors. One-tailed tests were used with the
`intent of most conservatively identifying factors as-
`sociated with the cortisol slope that might influence
`the relationship of cortisol slope and survival. The
`contributions of these variables to the cortisol slope–
`survival relationship were tested by entering each
`variable as a covariate with the cortisol slope in a
`separate Cox regression. Here, the predictive value
`of the cortisol slope was considered to remain after
`adjustment for these other variables only when it
`met the more stringent criterion of significance in a
`two-tailed test. The relative risk of medical variables
`that are common prognostic indicators in breast can-
`cer was evaluated by survival analyses performed by
`Cox regression. The Bonferroni correction was not
`applied to tests of the secondary hypotheses because
`they were exploratory in nature.
`
`RESULTS
`
`Flatter diurnal cortisol slopes predicted
`shorter subsequent survival times (Cox
`proportional hazards two-tailed P 4
`.0036; hazard ratio 4 464.9; 95% confi-
`dence interval [CI] 4 7.5–28 953.0). Al-
`though the survival analysis was con-
`ducted using the entire sample and
`cortisol slope as a continuous variable, for
`descriptive (not statistical) purposes, sub-
`jects were split into two equal groups
`based on the cortisol slope. Fig. 1, A,
`shows the mean cortisol levels at morn-
`ing, afternoon, and evening hours in these
`groups. Fig. 1, B, shows cortisol levels
`from four individual patients representing
`the various types of cortisol profiles that
`we observed. Kaplan–Meier survival
`plots of the two groups in Fig. 1, A, are
`
`shown in Fig. 2. The divergence in sur-
`vival as a function of slope emerged ap-
`proximately 1 year after cortisol assess-
`ment and extended at least 7 years after.
`Among patients split at the median corti-
`sol slope, 77% of those with flat rhythms
`died (average survival, 3.2 years; 95% CI
`4 2.5–3.9 years), while only 60% of
`those with steep rhythms died (average
`survival, 4.5 years; 95% CI 4 3.7–5.2
`years); survival plots of these groups di-
`verged significantly (log-rank, P 4 .016,
`two-tailed).
`Fig. 1, B, shows data from four pa-
`tients representing the various types of
`cortisol profiles observed. Only 37% of
`the patients had relatively normal diurnal
`rhythms, with cortisol concentrations that
`peaked at 0800 hours and levels that de-
`clined consistently thereafter. Peak con-
`centrations occurred later in the day for
`49% of the sample, while 14% had no
`apparent peak during the observed times
`of day. As suggested by the profiles
`shown in Fig. 1, B, in which all three
`patients with “aberrant” rhythms (open
`symbols) had lower morning and higher
`evening cortisol levels as compared with
`the “normal” diurnal rhythm (closed sym-
`bols), patients with flatter diurnal slopes
`had statistically significantly lower morn-
`ing and higher evening cortisol levels.
`The minimum cortisol concentration over
`all times of day was greater for patients
`with flatter slopes (Spearman R 4 .22, P
`4 .02).
`Patients with flatter slopes had fewer
`absolute numbers of circulating NK cells
`(Spearman R 4 −.25; P 4 .007). Sup-
`pression of NK cell activity was also evi-
`dent among patients with flatter slopes (R
`4 .17; P 4 .05). Table 2 presents the
`mean and SD for NK cell number and NK
`cell lytic units reflecting a 20% lysis for
`groups dichotomized at the median corti-
`sol slope for illustrative purposes as de-
`scribed in Figs. 1, A, and 2. Higher abso-
`lute NK cell counts predicted longer
`survival when used as a single indepen-
`dent variable in a Cox regression (P 4
`.037; two-tailed; hazard ratio 4 0.9977;
`95% CI 4 0.9956–0.9999) (n 4 95).
`However, when both cortisol slope and
`absolute NK cell counts were included in
`the regression, the slope emerged as a
`stronger predictor of survival (cortisol
`slope—two-tailed P 4 .013, hazard ratio
`4 339.1, and 95% CI 4 3.5–32 760.1
`versus NK cell counts—P 4 .069, hazard
`ratio 4 0.9980, and 95% CI 4 .9959–
`1.0002). NK cell activity was not associ-
`
`996 REPORTS
`
`Journal of the National Cancer Institute, Vol. 92, No. 12, June 21, 2000
`
`
`
`Downloaded from
`
`http://jnci.oxfordjournals.org/
`
` by guest on October 3, 2016
`
`Fig. 1. Panel A: Mean (bars show 95% confidence intervals) diurnal salivary
`cortisol levels at four times of day for two equal groups of patients split at the
`median diurnal cortisol slope (−0.091 log [mg/dL] per hour). The average value
`of the diurnal cortisol slope for the flat slope was -
`.040 log(mg/dL)/hour and for
`the steep slope group it was -
`.128 log(mg/dL)/hour. Panel B: Mean salivary
`cortisol levels at four times of day, with the diurnal cortisol slope shown in the
`legend for each plot. Data from four study participants with cortisol profiles that
`represent the several types of profiles observed. The data from patients who were
`
`in the flat-slope group after the dichotomous split used in Figs. 1, A, and 2 are
`shown in the open symbols. The data from one patient in the steep-slope group
`are shown with closed symbols. Only 37% of the patients had cortisol concen-
`trations that peaked at 0800 hours with levels that consistently declined there-
`after. This pattern is represented by the patient whose data are shown with closed
`circles. Another 49% of patients had peaks occurring later in the day (h and n),
`and 14% had no apparent peak during the observed times of day (s).
`
`psychosocial variables listed previously.
`There was a small but statistically signifi-
`cant association (P 4 .023) between me-
`tastasis in the chest wall or adjacent
`lymph nodes as opposed to visceral or
`bone metastases and steeper slopes
`(steeper slopes—i.e., their cortisol pro-
`files look more “healthy”). Patients who
`were taking megestrol had flatter slopes
`(P 4 .000). Patients who reported more
`nocturnal awakenings had flatter slopes
`(P 4 .003), and marital disruption was
`associated with flatter slopes (P 4 .040).
`Slope remained a statistically significant
`predictor of survival, even after adjust-
`ment for the effects of each of these vari-
`ables. No other variable related to disease
`or treatment was statistically significantly
`associated with the cortisol slope.
`As yet an additional check, we evalu-
`ated the predictive value of the common
`prognostic indicators in breast cancer
`(e.g., age at initial diagnosis, disease-free
`interval, and estrogen receptor status) in
`this sample by use of survival calculated
`in the more traditional manner, from the
`time of initial diagnosis. As expected, es-
`trogen receptor-negative tumor status and
`shorter disease-free interval predicted
`shorter survival measured from the date
`of initial diagnosis. However, once again,
`cortisol slope remained a statistically sig-
`nificant predictor of survival after adjust-
`ment for each of these variables.
`There were no differences in baseline
`
`Fig. 2. Kaplan–Meier survival curves for patients split into two equal groups at the median diurnal cortisol
`slope (−.091 log [mg/dL] per hour). This grouping was performed only to illustrate survival curves repre-
`senting patients with relatively steep versus flat cortisol slopes. The definitive survival analysis was con-
`ducted on the entire sample using the continuous variables of cortisol slope and survival time in a Cox
`regression. Patients with relatively flat cortisol slopes experienced shorter subsequent survival (Cox pro-
`portional hazards, two-tailed P 4 .0036). Among the patients split at the median cortisol slope, 77% of those
`with flat rhythms died, after surviving an average of 3.2 years (broken line). In contrast, 60% of the patients
`with relatively steep rhythms died, but they survived more than 1 year longer on average, with an average
`survival of 4.5 years (solid line). Survival plots of these groups diverged significantly (log-rank, two-tailed
`P 4 .016). Patients still living at the time of analysis are indicated by black vertical slash marks. The
`mumbers of living patients at each year mark are listed for the “flat-slope” and “steep-slope” groups whose
`survival curves are shown in the figure.
`
`ated with subsequent survival. Overall
`levels of cortisol (AUC) were not associ-
`ated with NK cell numbers or with their
`functional activity.
`
`As a check against possible confound-
`ing variables, we examined associations
`of the cortisol slope with the markers of
`disease status, medical treatments, and
`
`Journal of the National Cancer Institute, Vol. 92, No. 12, June 21, 2000
`
`REPORTS 997
`
`
`
`Downloaded from
`
`http://jnci.oxfordjournals.org/
`
` by guest on October 3, 2016
`
`Table 2. Mean (standard deviation) numbers and functional activity* of natural killer (NK) cells in
`groups of patients with metastatic breast cancer dichotomized at the median diurnal cortisol slope
`
`Cortisol slope log,
`mg/dL per h
`
`NK cells per mm3
`whole blood
`
`Lytic units,*
`20% lysis
`
`% lysis by effector-to-target cell ratio
`
`100 : 1
`
`50 : 1
`
`25 : 1
`
`12 : 1
`
`6 : 1
`
`Steep slope, ł−.091
`Flat slope, > −.091
`
`212 (131)
`164 (134)
`
`800 (473)
`851 (520)
`
`49 (22)
`44 (19)
`
`45 (21)
`39 (19)
`
`39 (19)
`33 (17)
`
`28 (14)
`25 (14)
`
`18 (9)
`16 (9)
`
`*Lytic units are corrected for NK cell numbers and are expressed as the number of NK cells required to
`kill 20% of the targets in a suspension of 107 cells per mm3.
`
`cortisol slopes between patients subse-
`quently randomly allocated to the psycho-
`social treatment group versus the control
`group. Moreover, when the association
`between the cortisol slope and survival
`was controlled for the effects of the ran-
`domized group assignment, the slope re-
`mained a statistically significant predictor
`of subsequent survival (P 4 .004, hazard
`ratio 4 459.6; 95% CI 4 7.1–29 799.7).
`As an additional check of the robustness
`of our finding, we examined the relation-
`ship of the cortisol slope to subsequent
`survival specifically among control group
`patients. The predictive effect of cortisol
`slope on survival held when analyzed in
`the control group only (Cox proportional
`hazards, P 4 .006, hazard ratio 4
`1376.9; 95% CI 4 7.8–242 263.5), de-
`spite the restricted statistical power that
`comes with a reduced sample size (n 4
`46).
`
`DISCUSSION
`
`For the first time, we document that
`variability in the diurnal cortisol rhythm
`is a statistically significant predictor of
`survival time with metastatic breast can-
`cer. Prior work (26) has shown that cir-
`cadian abnormalities have prognostic
`value in predicting the initial occurrence
`of breast cancer, in that those at high risk
`showed a markedly different circadian
`pattern for an array of hormones than did
`women at low risk. Moreover, other stud-
`ies (8,9) have associated circadian abnor-
`malities with later stages of cancer devel-
`opment or with prognostic indicators. We
`did not observe an association between
`the cortisol slope and several standard
`prognostic disease variables. Further-
`more, the cortisol slope predicted mortal-
`ity more than 12 months after assessment.
`This suggests that flattening of the corti-
`sol slope does not simply reflect preter-
`minal changes; rather, it is a relatively
`long-term prognostic indicator.
`Both mathematically and empirically,
`overall levels of cortisol measured by the
`
`AUC between 0800 and 2100 hours are
`functionally determined by a linear com-
`bination of the morning (0800) cortisol
`and the diurnal cortisol slope. Some read-
`ers will question whether high cortisol
`levels overall (AUC) would have been a
`better predictor of subsequent survival. A
`Cox regression was performed to test the
`predictive value of AUC on survival time.
`However, there was no association be-
`tween AUC and subsequent survival,
`highlighting cortisol slope, and not the
`morning cortisol level, as a unique indi-
`cator of survival in metastatic breast cancer.
`Lower NK cell numbers and NK cell
`activity were associated with flattened
`slopes. Effects of glucocorticoids on ac-
`tivity and trafficking of immune cells
`have been demonstrated (27), and NK cell
`counts may vary with cortisol levels (18).
`Patients with flatter diurnal patterns did
`not generally experience low levels of
`cortisol in the evening. It is possible that
`the absence of the typical evening cortisol
`nadir affected NK profiles and that the
`normal morning peak of NK cell count
`was disrupted in these patients. Thus,
`rather than reflecting a deficit of NK cells,
`our association of flat cortisol slope with
`lower NK cell counts may simply reflect
`an abnormality of NK cell trafficking sec-
`ondary to the disrupted cortisol rhythm.
`Nevertheless, lower numbers of NK cells
`were implicated in survival, and patients
`with flatter cortisol slopes had poorer NK
`cell activity. NK cells do kill several types
`of experimental tumors and may substan-
`tially defend against breast cancer pro-
`gression (28). Thus, immunosuppression
`related to flattened cortisol slopes may
`partly explain the survival effect. Reduc-
`tions in NK cell numbers have not been
`observed in disorders characterized by
`hypercortisolism (e.g., Cushing’s disease
`and depression) (29,30). Thus, it is not
`surprising that no association was found
`between overall cortisol levels (AUC) and
`NK cell numbers.
`The chest wall as the dominant meta-
`static site was associated with the cortisol
`
`slope but did not weaken the relationship
`between the slope and survival. In con-
`trast with chest metastases, more distant
`invasion such as visceral metastasis is a
`relatively end-stage event. Taken together
`with the lack of association between the
`slope and age, disease-free interval, estro-
`gen receptor status, time since recurrence,
`and Karnofsky rating, this finding sug-
`gests that the diurnal cortisol slope re-
`flects a different dimension of physi-
`ologic dysregulation related to earlier
`disease spread rather than to more severe
`changes associated with preterminal dis-
`ease states.
`There was no association between re-
`cent chemotherapy (i.e., treatment admin-
`istered within 2 months of data collection)
`and the diurnal cortisol slope, indicating
`that endocrine dysregulation is not simply
`a result of chemotherapy. The association
`between megestrol treatment and flatter
`slopes is worthy of further consideration,
`since loss of the cortisol rhythm may have
`consequences relevant
`to disease-
`resistance processes. Other hormonal
`therapies (e.g., tamoxifen therapy) were
`not associated with aberrations of the di-
`urnal cortisol rhythm.
`We explored the possibility that links
`between flattened cortisol rhythms and
`shorter survival were an epiphenomenon
`of other factors related both to cortisol
`variability and to survival. Results give
`tentative evidence that disruption of mari-
`tal ties through separation, divorce, or
`widowhood may be associated with the
`loss of normal diurnal cortisol variation.
`Flattened diurnal profiles may reflect the
`severe and chronic stress of losing marital
`support, which itself has been associated
`with poorer cancer outcomes (31). In-
`deed, other work (1,3) suggests that
`the physiologic burden of accumulated
`stressors on endocrine response systems
`may result in flattened cortisol rhythms
`and may have long-term health conse-
`quences.
`Cortisol slope was not related to pain
`or depression, and neither of these factors
`predicted survival. Nevertheless, persis-
`tent hypercortisolism could have trig-
`gered psychophysiologic responses that
`perpetuated the abnormality. For ex-
`ample, nocturnal hypercortisolism is as-
`sociated with sleep disturbance (32). In
`the current sample, flatter slopes were as-
`sociated with more nocturnal awakenings.
`The failure of a normal nadir at bedtime
`may have disrupted sleep, a likely cause
`of further circadian disruption.
`
`998 REPORTS
`
`Journal of the National Cancer Institute, Vol. 92, No. 12, June 21, 2000
`
`
`
`Downloaded from
`
`http://jnci.oxfordjournals.org/
`
` by guest on October 3, 2016
`
`These data were collected, not only to
`test the relationship of the diurnal cortisol
`slope to survival but also to examine the
`hypothesis that psychosocial treatment
`will increase survival time. Although
`participants were assigned to either the
`treatment or the control group only after
`cortisol data were collected, group assign-
`ment constitutes a “nuisance” parameter
`in this analysis. Indeed, in the context of
`this randomized trial, any variability in
`survival time introduced by the interven-
`tion would tend to cloud relationships be-
`tween pre-existing physiologic markers,
`such as the baseline cortisol slope, and
`subsequent survival. Our finding that the
`cortisol slope remained a statistically sig-
`nificant predictor of survival, both after
`adjustment for assignment to the treat-
`ment or control group and when exam-
`ined among control patients only, sug-
`gests that
`it
`is a robust prognostic
`indicator of survival time.
`Altered cortisol rhythmicity in cancer
`has been associated with other circadian
`abnormalities as well as poor prognosis
`(8), making it possible that our results re-
`flect circadian abnormalities extending
`beyond the adrenocortical axis. Aberrant
`cortisol rhythms may be associated with
`hypersensitivity to stress (3) or may dis-
`rupt other endocrine and immune rhythms
`(18). These effects in combination with
`the psychologic stress of cancer may
`cause deterioration of disease-resistance
`capabilities (1). We are examining this
`possibility by studying stress reactivity as
`well as the rhythms of other physiological
`end points in these patien