Fax: (215) 349-5067
Is screening effective in detecting untreated psychiatric disorders among newly diagnosed breast cancer patients?
Article first published online: 11 OCT 2011
Copyright © 2011 American Cancer Society
Volume 118, Issue 10, pages 2735–2743, 15 May 2012
How to Cite
Palmer, S. C., Taggi, A., DeMichele, A. and Coyne, J. C. (2012), Is screening effective in detecting untreated psychiatric disorders among newly diagnosed breast cancer patients?. Cancer, 118: 2735–2743. doi: 10.1002/cncr.26603
- Issue published online: 3 MAY 2012
- Article first published online: 11 OCT 2011
- Manuscript Accepted: 27 JUL 2011
- Manuscript Revised: 25 JUL 2011
- Manuscript Received: 30 MAR 2011
- breast neoplasms/psychology;
- psychotropic drugs/therapeutic use;
A key purpose of routine distress screening is to ensure that cancer patients receive appropriate mental health care. Most studies validating screening instruments overestimate the effectiveness of screening by not differentiating between patients with untreated disorders and patients who are already being treated. This study adopts the novel strategy of evaluating the effectiveness of screening after correcting for disorder for which treatment is already being provided.
A total of 437 recently diagnosed breast cancer patients received in-clinic distress screening and telephone-based psychiatric interviews. Analyses were conducted using receipt of psychotropic medication for mental health difficulties in the context of a psychiatric disorder as a proxy for identification and treatment.
Rates of elevated distress (33%), major depressive disorder (8%), minor depression (6%), dysthymia (2%), or generalized anxiety disorder (3%) were similar to those in other samples. Thirty-six percent of patients received psychotropic medication around the time of cancer diagnosis, including 64% of those with a current psychiatric diagnosis. Although 39% of patients with elevated distress had a psychiatric disorder, the positive predictive value of screening fell to 15% for an untreated psychiatric disorder and 6% had untreated depression.
Given the high rates of existing treatment, screening may not be efficient for identifying untreated disorder. Almost two-thirds of patients with treated disorders remain symptomatic. Use of symptom scales might reasonably be expanded to surveillance of treatment response or ruling out disorder. Substantial resources would likely be required to coordinate or manage psychiatric care among patients, as would a willingness to intervene in existing relationships with other providers. Cancer 2011. © 2011 American Cancer Society.
A National Institutes of Health Consensus Conference Statement1 and an Institute of Medicine report2 have identified detection and management of distress and psychiatric disorder among cancer patients as a priority. Screening is often proposed as a first step in providing appropriate psychosocial and mental health care, and as a means of reducing morbidity and improving quality of life.3-5 Evidence-based guidelines for managing distress, however, are still being developed.6 Moreover, a considerable proportion of positive screens for distress represents psychiatric disorder, typically 30%-45%.7, 8 Such disorders tend to be chronic, episodic, and recurrent. They require more intensive care and longer follow-up, but have the advantage of an agreed-upon nosology and well-defined, empirically based treatment guidelines. Evaluations of screening instruments, therefore, typically make gold standard comparisons against validated diagnostic interviews for psychiatric disorders.9, 10
Questions have been raised about the appropriateness of particular screening instruments for specific subgroups of cancer patients,11, 12 though most screening instruments appear to have moderate performance compared with diagnoses based on psychiatric interviews.10 Even the best performing instrument cannot be used diagnostically, hence screening is best construed as an initial step in a process in which positive screens are followed by a diagnostic interview.13 Given the modest prevalence of psychiatric disorders in most cancer populations, however, most positive screens will be found to be false positives, rendering screening potentially inefficient.8
Routine screening programs assume that screening provides a means of uncovering distress and psychiatric disorder that would not otherwise be identified and treated.14, 15 In clinical practice, the interest is not in detecting patients already being treated, but in those who would otherwise be missed. A shortcoming of existing studies of the accuracy and yield of screening tools in cancer care is that they are almost always conducted with all available patients, including patients who have already been recognized and are receiving treatment for psychiatric disorder, inflating estimates of the number of patients in need of treatment that screening can yield. A recent systematic review of the diagnostic accuracy of screening instruments for depression found only 8 of 197 (4.1%) unique reports from 17 systematic reviews and meta-analyses specifically excluded patients already diagnosed or being treated for depression.16 Only 2 of these 8 reports concerned cancer patients,11, 17 both of which were from the same sample, and these simply excluded diagnosed or currently treated depression without determining the effect of doing so on screening performance.
Patients with a psychiatric disorder who are already in treatment may differ from those who are not by having a more conspicuous disorder or a more positive attitude about treatment.18 Patients for whom screening represents the only means of identifying psychiatric morbidity may, therefore, be more difficult to diagnose and require additional resources to overcome resistance to treatment, whereas patients who are already in treatment but remain sufficiently symptomatic to screen positive may require improved surveillance and greater coordination of care with outside providers. Routine care for depression in the community tends to be inadequate, with low adherence, high rates of unsupervised discontinuation of treatment, and a lack of scheduled follow-up visits.19, 20 These problems suggest the value of surveillance of existing treatment, with a recognition that the clinical resources required would differ substantially from those required if the goal were simply initiation of new treatment based on screening.
We sought to evaluate the performance of screening among newly diagnosed breast cancer patients, comparing screening results to a validated diagnostic interview, first, as is typically done, ignoring whether patients were already in treatment, and, second, taking existing psychotropic treatment into account. We chose newly diagnosed patients because there is evidence that distress is most prominent early in the cancer trajectory21-26 and predicts long-term adaptation.27 Our objectives were to (1) evaluate the ability of a screening instrument to detect depression and other psychiatric disorders among a sample of patients in early treatment for breast cancer, and (2) determine how screening performance is affected when existing treatment for a disorder is taken into account.
MATERIALS AND METHODS
Recently diagnosed breast cancer patients seeking treatment at a comprehensive cancer center were eligible to participate. Inclusion criteria included age 18-85 years, ability to understand and speak English, and enrollment within 5 months of diagnosis and before chemotherapy. Exclusion criteria included current recurrence of cancer within 5 years, current substance abuse, and history of bipolar or psychotic disorders.
Potential participants were approached after their first medical oncology or postoperative surgical visit and asked to participate in a longitudinal study of the psychosocial impact of breast cancer and its treatment. Those expressing interest provided informed consent. Participants were given questionnaires to be completed and returned by mail within 1 week. Questionnaires included the distress screening measure, demographics, and additional psychosocial measures. Patients who did not return their questionnaires within 2 weeks received a telephone call to prompt return of the packet. A semistructured mental health telephone interview was scheduled within 2 weeks, and only took place after receipt of screening materials. During the interview, the purpose of the study was reviewed and participant questions answered. Interviews were recorded for reliability purposes. Ethics approval was granted by the University of Pennsylvania Institutional Review Board and the Clinical Trials and Scientific Monitoring Committee of the Abramson Cancer Center.
Demographics included age, race, education, marital status, and household income.
Prescription of psychotropic medication for psychiatric disorder was assessed by interview. We were concerned about patients reporting medications received for non–mental health reasons (eg, hot flashes, pain), so patients were asked if they had been “given a prescription in the past 6 months for problems with stress, emotions, nerves, drugs, alcohol, or mental health.” Medication names and dosages were reported by direct patient inspection of available bottles, and medications were categorized as “antidepressant,” “anxiolytic,” or “other” (eg, hypnotic). For current analyses, only anxiolytic and antidepressant prescriptions were examined.
Chart abstraction/clinical characteristics
Cancer treatment information, disease staging, diagnosis date, and occurrence of metastases/recurrences were abstracted from charts by a trained medical abstractor.
The 25-item28 Hopkins Symptom Checklist29 (HSCL-25) was used to screen for psychological distress. The items of the HSCL-25 either overlap with or have inconsequential differences from items on various versions of the Brief Symptom Inventory.30-32 As a group, this family of instruments is widely used among cancer patients.33-36 Hough and colleagues28 found that the HSCL-25 was comparable or superior to the CES-D37 in detecting psychiatric disorder among general medical patients. The HSCL-25 has demonstrated reliability.38-40 In the current sample, Cronbach's alpha coefficient was good (0.93).
Diagnoses were obtained using the Structured Clinical Interview for DSM-IV/NP (SCID),41 a semistructured interview based on DSM-IV.42 Modules for current major depressive disorder (MDD) and lifetime history of past MDD, dysthymic disorder, and generalized anxiety disorder (GAD) were administered via telephone by trained research staff blind to participant responses on questionnaires. Studies have shown good concordance between telephone and face-to-face diagnostic interviews.43-46 Interrater reliability was assessed on 10% (n = 44) of interviews for MDD and past MDD diagnoses. Agreement at the symptom level was high (91%), and reliability was adequate (kappa = 0.70-0.85).
Primary Data Analysis
Using the standard cut-point of 44 on the HSCL-25, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for the presence of MDD and a category combining “any psychiatric disorder,” which included MDD, minor depression, dysthymia, and GAD. Confidence intervals (CIs) for proportions were constructed using the modified Wald method.47 Next, patients with a psychiatric diagnosis that was already being treated pharmacologically were removed from the sample, and performance characteristics were recalculated to estimate performance for the untreated patients who would be of greatest interest in clinical practice. All hypothesis tests were 2-tailed.
Characteristics of Sample
A total of 652 patients were approached to participate, and 588 consented (90%). Of these, 535 (91%) contributed questionnaire or interview data, and 437 (82%) contributed both and are included in the current analysis for an overall participation rate of 67%. Patients contributing both questionnaire and interview data did not differ from those contributing one but not the other for presence of any psychiatric disorder, psychotropic medication, distress, cancer stage, race, income, or marital status (P ranges .10-.80). Table 1 provides demographic information for the sample.
|Age, y||Mean ± SD, 54.2 ± 11.7 (range, 27-83)|
|Time since diagnosis, d||Mean ± SD, 56.1 ± 33.1 (range, 0-160)|
|College graduate or higher||55%|
|At least high school||97%|
|Stage at diagnosis|
|Positive ER status||80%|
|Positive PR status||70%|
|Positive Her2/neu status||15%|
Distress, Psychiatric Diagnosis, and Psychotropic Medication
Overall, psychological distress was moderate (mean ± standard deviation, 40.2 ± 10.7) and significantly below the cut-point of 44 indicating “elevated distress” (t(436) = −7.3; P<.001). However, 33% of the sample (95% CI, 28%-37%; n = 142) scored above the established cut-point.
Criteria for current MDD were met in 8% of the sample (95% CI, 6%-11%; n = 36), whereas 2% (95% CI, 0.7%-3%; n = 7) met criteria for dysthymia and 3% (95% CI, 2%-5%; n = 14) met criteria for GAD. Minor depression research criteria were met in 6% (95% CI, 4%-9%; n = 26). Criteria for at least 1 of these psychiatric disorders were met by 17% of patients (95% CI, 14%-21%; n = 74). Approximately 20% (95% CI, 16%-24%; n = 87) of patients reported a past history of MDD on the SCID.
At interview, 36% (n = 158) of patients had a prescription for a psychotropic medication. Twenty percent (n = 88) of patients had a prescription for an antidepressant, 26% had a prescription for an anxiolytic (n = 112), and 10% (n = 42) had a prescription for both. Almost two-thirds (64%; n = 47) of patients with a current psychiatric diagnosis were receiving psychotropic medication (X(1)2 = 28.89; P ≤. 001), with 46% (n = 34) receiving antidepressants and 42% (n = 31) receiving anxiolytics. With respect to specific diagnoses (Table 2), more than three-quarters of patients with current MDD (n = 28) were receiving psychotropic medication, with similar proportions of antidepressant (n = 20) or anxiolytic (n = 19) prescriptions. Minor depression and dysthymia were less likely to have been treated, whereas GAD was treated at a rate of 71% (n = 10), with antidepressants (n = 7) and anxiolytics (n = 7) being equally likely to have been prescribed. Close to half of patients with a history of MDD but without current disorder (n = 27) were receiving treatment. However, prescriptions were nonspecific, and more than one-quarter (28%; n = 81) of patients with neither current nor past disorder received psychotropic medication, accounting for 53% of all psychotropic prescriptions.
|MDD||78% (n = 28)a||56% (n = 20)a||53% (n = 19)a|
|Minor depression||50% (n = 13)||39% (n = 10)a||35% (n = 9)|
|Dysthymia||43% (n = 3)||43% (n = 3)||14% (n = 1)|
|GAD||71% (n = 10)a||50% (n = 7)a||50% (n = 7)a|
|Past MDD, no current diagnosis||48% (n = 27)a||30% (n = 17)a||34% (n = 19)a|
|No current or past diagnosis||27% (n = 81)||12% (n = 36)||20% (n = 59)|
Detecting Psychiatric Disorder Through Screening
Elevated distress on the HSCL-25 related significantly to the presence of current MDD (X(1)2 = 67.88, P ≤ .001). This association translated into modest screening efficiency (Table 3). Although only 2 of 36 patients with current MDD were missed, 76% of those who screened positive did not have MDD (n = 109), and, reflecting the overall prevalence of MDD in this sample, a patient screening positive would be 3.2 times more likely to not have MDD than to have MDD. A negative screen, however, was very efficient in ruling out MDD (NPV = 0.99).
|Sensitivity (95% CI)||Specificity (95% CI)||PPV (95% CI)||NPV (95% CI)||True Positive, n||False Positive, n||True Negative, n||False Negative, n|
|MDD||0.94 (0.82-0.99)||0.73 (0.72-0.73)||0.24 (0.21-0.25)||0.99 (0.98-1.0)||34||109||292||2|
|Any disorder||0.74 (0.65-0.82)||0.76 (0.74-0.78)||0.39 (0.34-0.43)||0.94 (0.91-0.97)||55||88||275||19|
|Untreated MDD||0.88 (0.53-0.98)||0.73 (0.72-0.73)||0.06 (0.04-0.07)||1.0 (0.99-1.0)||7||109||292||1|
|Any untreated disorder||0.56 (0.38-0.72)||0.76 (0.75-0.77)||0.15 (0.10-0.19)||0.96 (0.94-0.97)||15||88||275||12|
|Untreated MDDa||0.88 (0.47-0.99)||0.78 (0.77-0.79)||0.11 (0.06-0.12)||1.0 (0.98-1.0)||7||59||212||1|
|Any untreated disordera||0.56 (0.38-0.73)||0.80 (0.78-0.82)||0.23 (0.15-0.30)||0.94 (0.92-0.97)||15||51||201||12|
HSCL-25 scores related significantly to the presence of any psychiatric disorder (X(1)2 = 70.03; P ≤ .001). However, the HSCL-25 missed 26% (n = 19) of patients with a current disorder, whereas 24% (n = 88) of patients with no current disorder screened positive, accounting for 61% of positive screens.
Detecting Untreated Psychiatric Disorders
When patients with treated MDD were removed from the analysis, a positive screen for distress remained significantly related to the presence of MDD (X(1)2 = 14.05; P < .001). Screening missed only 1 of 8 patients with MDD; however, 95% (n = 109) of those patients receiving positive screens (n = 115) did not have MDD. Moreover, although 28% of patients (n = 116) screened positive, only 6% (n = 7) had MDD that had not already been identified and treated. (A more conservative approach would be to exclude as treated those patients with MDD who were receiving antidepressant medication only rather than antidepressants or anxiolytics combined, because antidepressants are the preferred medication for MDD. Results of analysis using this method are quite similar to those removing individuals receiving either medication [X(1)2 = 32.6; P < .001; sensitivity = 0.94; specificity = 0.72; PPV = 0.12; NPV = 1.0]).
When all patients with treated disorder were removed from analysis, a positive screen for distress still significantly predicted presence of any untreated psychiatric disorder (X(1)2 = 12.68; P < .001). However, a positive screen for distress missed 44% (n = 12) of untreated psychiatric disorders (n = 27), whereas 24% (n = 88) of patients without a disorder continued to screen positive, accounting for 85% of all positive screens.
To examine whether high rates of prescription of psychotropic medication in this sample reflected potentially appropriate maintenance or prophylactic treatment of past major depressive disorder, we analyzed data excluding patients with current treated disorder and/or past MDD receiving current treatment. For untreated MDD, results were similar to those given in Table 3 (sensitivity = 0.83; specificity = 0.78; PPV = 0.06; NPV = 1.0), as were those obtained considering presence of any untreated disorder (sensitivity = 0.57; specificity = 0.79; PPV = 0.13; NPV = 0.97).
Finally, to approximate more closely what would happen in clinical practice if decisions about screening were based only on information at hand (eg, whether or not a patient is receiving treatment), we reanalyzed data excluding all patients receiving psychotropic medication regardless of presence of disorder. Results were substantially similar using the HSCL-25 for both detection of untreated MDD (X(1)2 = 18.69; P < .001; sensitivity = 0.88; specificity = 0.79; PPV = 0.11; NPV = 1.0) and any untreated current disorder (X(1)2 = 16.84; P < .001; sensitivity = 0.57; specificity = 0.80; PPV = 0.23; NPV = 0.94).
When evaluated against a diagnostic interview with no adjustment for prior identification, the HSCL-25 performed comparably to other instruments.10 A negative screen was an accurate indicator of the absence of psychiatric disorder. However, the number of true positive cases was dwarfed by false positives. Moreover, when patients with a psychiatric disorder that had already been identified and treated were excluded from analyses, the ratio of true positive identifications relative to false positives was further reduced. Given that sensitivity varies only slightly and specificity not at all, this effect largely results from the lower prevalence rate for untreated patients relative to the entire sample. That is, removal of treated individuals reduces the prevalence of disorder from approximately 16% (74/437) to 7% (27/390). Our results raise questions about the efficiency of routine screening as a means of identifying psychiatric disorder that would otherwise go untreated, and point to the need for future studies validating screening instruments to take existing treatment into account.
The prevalence of distress in our sample was comparable to what is found in other samples.48, 49 Prevalence estimates of MDD or other disorders in past research vary widely, largely due to methodological differences.50 However, rates that are lower and similar to those in the current study are obtained when research interviews, rather than self-report instruments, are used to determine caseness.49, 51, 52 An expert consensus statement has concluded that rates of psychiatric disorder among cancer patients are comparable to those found in other medical populations, but higher than the general population.53
The high rates of prescription of psychotropic medication were striking and need to be interpreted in the context of escalating rates of prescription in the general population.54, 55 In a previous study of longer-term breast cancer survivors,7 we found similar prescription rates, but that sample was heterogeneous with respect to time because diagnosis and 66% of the sample had been diagnosed 2 or more years previously. The present sample was less than 2 months from diagnosis, on average. It is, however, unclear whether psychotropic medication had been prescribed after diagnosis of cancer or before diagnosis.
There are few data concerning prescription of antidepressants in cancer care, but there are suggestions that rates are high relative to the prevalence of MDD. Ashbury et al56 found that 19.2% of breast cancer patients received an antidepressant prescription in the 2 years following diagnosis, and that prescription correlated to receipt of pain medication. This likely underestimates prescription, because Ashbury et al included only those prescriptions obtained or recorded through oncology services, and many patients receive prescriptions through primary care.57 A study of terminally ill cancer patients found that 40% had an antidepressant prescription, although the prevalence of MDD was only 17%.58 Previous work has suggested that repeated contact with non-mental health specialty physicians increases rate of prescription of psychotropic medications, but not necessarily quality of care.59 Future reports from our dataset will examine type of provider and changes in prescription rates up to a year after diagnosis.
Psychotropic prescriptions appeared nonspecific, and more than one-quarter of patients with neither current nor past psychiatric diagnosis were receiving medication for “stress, emotions, nerves, drugs, alcohol or mental health” around cancer diagnosis. Another study7 suggested that nonspecific prescription increases over time, rather than plateauing. In that study, 48% of long-term breast cancer survivors with neither current nor past psychiatric diagnosis received an anxiolytic or antidepressant. Clearly, there is a need to examine whether this leads to improved outcomes, though the literature on treatment of subsyndromal depression suggests that this is unlikely.60, 61
We found that the proportion of patients with unidentified and untreated psychiatric disorder was considerably less than the proportion of those who were receiving treatment but remaining syndromal. If replicated, our results may suggest that focusing efforts away from detecting untreated disorder toward surveillance of clinical response of already treated patients and using this information to improve quality of care may be worthwhile. Instruments that have been used for screening have been used as symptom scales to monitor treatment outcomes,62-64 and there have been demonstrations that improving consistency and quality of care provided to already identified patients with MDD would have a larger impact on a population basis than would increasing the number of patients introduced to treatment.65 Other data suggest that among cancer patients, as few as 15% of those with MDD receive treatment at a therapeutic level.66 The quality of routine depression care in the community is generally poor, particularly in general and tertiary nonpsychiatric medical settings,67, 68 and there may be particular problems ensuring high quality care within the competing demands of cancer. In addition, assuming responsibility for insuring appropriate care of patients already receiving prescriptions would probably require different staffing, resources, and relations with prescribing providers than those currently present in oncology care.
Other data suggest that it may be challenging to engage patients who have untreated psychiatric disorder in appropriate treatment. A recent study examining screening for “high-risk” medical patients69 found that of the 1687 patients invited for screening, only 71 were identified with MDD, and 36 were already in treatment. Fourteen of the remaining 35 patients refused treatment, and another 4 did not show up for an appointment. Ultimately, the number needed to be invited for screening in order for 1 previously unidentified depressed patient to receive treatment was 118. It seems likely that negative attitudes toward depression treatment are at least partially responsible for patients having their depression unidentified,18, 70 and will further complicate engaging them in treatment.
This study had the benefit of a high participation rate, a relatively large sample, and use of a validated psychiatric interview. It is limited by having been conducted at a single institution with a well-resourced population, and we cannot evaluate the degree to which results generalize to lower resourced settings or populations. The low prevalence of psychiatric disorder led to a small number of patients with any given disorder, particularly when examining untreated disorder, though confidence intervals take this into account. Our results likely underestimate the degree to which psychiatric disorder is already identified, because we do not have data concerning nonpharmacologic interventions or whether patients may have been offered but declined treatments. Therefore, our estimates of the effects of taking existing treatment into account should be seen as conservative in that consideration of psychotherapy and other psychosocial interventions would have increased the number of patients classified as receiving treatment.
We ascertained prescriptions via self-reporting. Although there are limitations to self-reporting, patients were asked to have all medication bottles available during the interview, and we inquired specifically about medications provided for “problems with stress, emotions, nerves, drugs, alcohol, or mental health.” However, we lacked systematic data about timing, indications, and sources of prescriptions, as well as adherence. It remains possible that some prescriptions were made for off-label indications. However, our interviews specified prescriptions for mental health–related difficulties, we ignored prescriptions for non-antidepressant or anxiolytic medications (eg, Zolpidem), and significant relationships between psychiatric diagnosis and receipt of prescription suggest that this was not a primary reason for prescriptions. It is possible that some prescriptions were missed through self-reporting, though rectifying this would likely strengthen rather than attenuate results. Finally, we did not take into account nonpharmacological treatment, treatment refusal, or adherence. Although some psychotherapy for depression is evidence-supported,71 the availability of empirically supported psychotherapeutic interventions among patients drawn from nonpsychiatric medical settings is likely to be low.72
In conclusion, our data indicate that screening may be less efficient as a means of identifying untreated psychiatric disorder among newly diagnosed breast cancer patients than would be inferred from estimates in the current literature of the efficiency of screening. Given the high proportion of patients with disorders who are treated without adequate remission of symptoms, and the difficulties likely to be experienced engaging nonidentified patients in treatment, screening measures might be better used as an initial step in monitoring the effectiveness of ongoing treatment, ruling out disorder,8 or as a prompt for improving communication surrounding psychosocial issues rather than a means of identifying individuals with untreated disorder. Clinicians in cancer care settings, however, will require significant resources if they are to effectively coordinate or manage psychiatric care among patients, in addition to having a willingness to intervene in existing treatment relationships with other providers.
Supported by Grant Number R01MH63172 from the National Institute of Mental Health and the National Cancer Institute.
CONFLICT OF INTEREST
The authors made no disclosures.
- 1National Institutes of Health State-of-the-Science Conference Statement. Symptom management in cancer: pain, depression, and fatigue. July 15-17, 2002. J Natl Cancer Inst. 2004; 32: 9-16., , .
- 2Adler NE, Page AEK, eds. Cancer Care for the Whole Patient: Meeting Psychosocial Health Needs. Washington, DC: The National Academies Press; 2007
- 3The NCCN guideline for distress management: a case for making distress the sixth vital sign. J Natl Compr Canc Netw. 2007; 5: 3-7., .
- 6Implementation of NCCN distress management guidelines by member institutions. J Natl Compr Canc Netw. 2007; 5: 99-103., .
- 28Comparison of psychiatric screening questionnaires for primary care patients. Final report for NIMH Contract 278-0036 (DB). Bethesda, MD: National Institute of Mental Health; 1982., , , et al.
- 30Assessment of psychiatric-illness severity by family physicians. Soc Sci Med. 1978; 12: 45-47., , , et al.
- 35Effects of tamoxifen and exemestane on cognitive functioning of postmenopausal patients with breast cancer: results from the neuropsychological side study of the tamoxifen and exemestane adjuvant multinational trial. J Clin Oncol. 2010; 28: 1294-1300., , , et al.
- 37The CES-D scale: a self-report depression scale for research in the general population. Appl Psychol Meas. 1977; 1: 385-401..
- 41Structured clinical interview for DSM-IV-TR Axis I disorders, research version, non-patient edition. New York: Biometrics Research, New York State Psychiatric Institute; 2001., , , et al.
- 42American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders, 4th Edition. Washington, DC: American Psychiatric Association; 1994.
- 47Approximate is better than “exact” for interval estimation of binomial proportions. Am Stat. 1998; 52: 119-126., .