Mental disorders contribute substantially to all-cause mortality.Reference Prince, Patel, Saxena, Maj, Maselko and Phillips1 However, the mechanisms underlying these associations remain unclear. Historically, research has prioritised depression and cardiovascular mortality, although similar associations have been found for other disorders such as schizophreniaReference Heila, Haukka, Suvisaari and Lonnqvist2 and bipolar disorder.Reference Osby, Brandt, Correia, Ekbom and Sparen3 Furthermore, depressive symptoms have been found to be associated with an increased risk of a range of causes of death, and strengths of association with cancer mortality are as strong as those with cardiovascular mortality.Reference Mykletun, Bjerkeset, Dewey, Prince, Overland and Stewart4 Treating comorbid psychiatric conditions may improve the likelihood of cancer survivalReference Chan, Ahmad, Yusof, Ho and Krupat5 because depression has been specifically associated with decreased overall survival after cancer diagnosis, although anxiety symptoms may be associated with increased survival.Reference Cuijpers, Vogelzangs, Twisk, Kleiboer, Li and Penninx6 These symptoms might modify help-seeking behaviours and other lifestyle factors that subsequently influence time to detection and diagnosis; this would account for mortality associations because cancer survival is strongly associated with extent of disease (EOD) at diagnosis.
Investigations of the association between mental disorder and cancer presentation present substantial logistic and methodological challenges primarily because obtaining sufficient numbers of people with both disorders requires a large source sample. Most studies have relied on retrospective evaluation of prior mental health in people who have received a cancer diagnosis,Reference Ristvedt and Trinkaus7–Reference Burgess, Potts, Hamed, Bish, Hunter and Richards11 with high potential for recall bias. The few studies with previously recorded data on mental health have relied on primary care recordsReference Robertson, Campbell, Smith, Donnan, Sullivan and Duffy12 or on relatively restrictive diagnostic criteria,Reference Desai, Bruce and Kasl13 both of which may underestimate morbidity. We sought to overcome these difficulties through carrying out a record linkage between a large cross-sectional survey whose participants had been systematically screened for anxiety and depressive symptoms, and a national cancer registry with information on EOD at diagnosis. Our hypothesis, given earlier mortality findings, was that beyond-local EOD at cancer diagnosis would be associated positively with previous depression, would be negatively associated with previous anxiety and would show no elevation or reduction in people with both states.
Method
The Cancer Registry of Norway (CRN; www.kreftregisteret.no) was established in 1951 and, from that point onward, all hospitals, laboratories of pathology and general practitioners (GPs) have been legally required to report neoplasms to the CRN.Reference Larsen, Småstuen, Johannesen, Langmark, Parkin and Bray14 The CRN discriminates between pre-malignant and malignant tumours, classifying diagnoses according to the ICD-715 system as well as coding the EOD at diagnosis (as local, regional, distant or unknown). Beyond prevention efforts, these screening programmes gather data about incidence and mortality. The analysis described here was carried out by linking the CRN database with the second Health Study of Nord-Trøndelag County (HUNT-2). HUNT-2 has been described in detail previously;Reference Holmen, Midthjell, Krüger, Langhammer, Holmen and Bratberg16 however, in summary it was a cross-sectional survey carried out between 1995 and 1997 of adult residents in one of Norway's 19 counties, containing 3% of the national population. Based on updated population register lists, all inhabitants (N = 93 138) in Nord-Trøndelag County aged ≥20 years received a mailed questionnaire and were invited to a clinical examination, resulting in 65 648 participants (71% response rate). The clinical examination was wide-ranging, including biometric measurements and blood assays, as well as the administration of the Hospital Anxiety and Depression Scale (HADS) described below. At the time of the survey, the population of Nord-Trøndelag County was relatively rural and stable with a net migration of 0.3% per year, and also relatively homogenous with less than 3% of the population identifying as non-White. Non-responders were younger and more likely to be male than participants.Reference Holmen, Midthjell, Krüger, Langhammer, Holmen and Bratberg16
The linked database comprised first CRN registrations after the time of the HUNT-2 survey (i.e. those with prior registrations were not included). For solid (i.e. rather than haematological) malignant tumours the EOD at diagnosis is, at the CRN, categorised as local (within the organ), loco-regional (spread outside the organ limits or regional metastases) and distant metastases. For this analysis, EOD was defined as a binary dependent variable: local EOD and beyond-local EOD. After initial descriptive analyses, unknown or missing EOD at diagnosis was considered as an exclusion criterion; haematological cancers were not included because of EOD non-applicability. Basal cell skin carcinoma was also excluded in accordance with standard practice for CRN output, on the basis that beyond-local spread is not a feature of this condition.
The HADS is a self-report questionnaire comprising 14 four-point items (seven for anxiety symptoms (HADS-A) and seven for depressive symptoms (HADS-D)) enquiring about symptoms during the past week.Reference Zigmond and Snaith17 The scale has been widely used and was specifically developed to detect anxiety and depressive symptoms in the context of somatic illness (hence, omitting somatic items, sleeping difficulties or appetite disturbance), with good case-finding properties in primary care and hospital settings (a cut-off score of 8 on each subscale found to give an optimal balance between sensitivity and specificity, both at about 0.80, for depression and anxiety diagnoses).Reference Bjelland, Dahl and Haug18 For brevity, case-level anxiety symptoms and depressive symptoms are hereafter referred to as ‘anxiety’ and ‘depression’, respectively. As in previous analyses using this measure and these cut-off points, three groups were identified: anxiety only, depression only and comorbid anxiety and depression. All three were compared separately to a fourth reference group with no disorder.Reference Roness, Mykletun and Dahl19
Specific covariates were chosen a priori from the source databases as potential confounders because they were known to be associated with common mental disorder symptoms and also had at least a potential influence on pathways to cancer presentation and diagnosis. These comprised the following: demographic factors of age, gender and education (categorised into three groups: compulsory only, college level and university level); smoking status (binary variable: current smoker or not), applied as a covariate as an indicator of health risk behaviour; living alone (binary variable), applied as a covariate assuming that participants with more social support might present at an earlier stage; level of physical health, quantified by number of self-reported conditions (angina pectoris, asthma, diabetes, epilepsy, hypertension, myocardial infarction, musculoskeletal disorder, respiratory disease, stroke and thyroid disease); type of first lifetime cancer (ICD-7 codes), as recorded by the CRN between 12–36 months after the HUNT-2 survey (type of cancer was considered as a covariate in this analysis on the basis that it is a strong determinant of EOD and that there is some evidence for associations between mental disorders and certain cancer types);Reference Fernando and Ince20,Reference Oerlemans, van den Akker, Schuurman, Kellen and Buntinx21 and finally, reported recent (during the past 12 months) physician contact before the HUNT-2 survey was available on a slightly smaller sample and was entered as a covariate in exploratory secondary analyses as it could potentially represent a factor lying on the causal pathway between mental state in HUNT-2 and cancer EOD at subsequent diagnosis (assuming that individuals in more regular physician contact may be more likely to receive an earlier diagnosis).
Statistical analysis
Statistical analyses were carried out using SPSS version 17.0 for Windows. The analysed sample was restricted a priori to HUNT-2 participants who had received a cancer diagnosis within 1–3 years of participation, having had no previous cancer diagnosis. Diagnoses within 1 year were excluded because of the possibility that some common mental disorder symptoms might be early manifestations of the cancer itself. Diagnoses beyond 3 years were excluded because of the diminishing likelihood that common mental disorder symptoms recorded at HUNT-2 participation would reflect symptoms around the time between first manifestations of cancer and diagnosis. Sample size was determined simply by the overlap between the two data-sets. After description of the sample, associations between case-level anxiety or depression and beyond-local EOD at cancer diagnosis (restricting to the subsample with data on EOD) were investigated in unadjusted analyses followed by logistic regression models to assess the influence of potential confounding factors. As stated above, adjustment for recent GP contact was treated as an exploratory secondary analysis. Statistical significance was set at P < 0.05.
Ethical approval
HUNT and the current project were approved by the National Data Inspectorate and the Board of Research Ethics in Health Region Midt-Norge of Norway (reference number 4.2008.2489). All individuals gave their written informed consent upon HUNT participation.
Results
In total, 681 first lifetime cancer diagnoses were made in HUNT-2 participants between 1–3 years after the time of participation. EOD at diagnosis was recorded in 459; of these, 407 (59.8% of the 681) had data on HADS subscales, comprising the analysed sample. Participants with missing HADS data were older (mean difference, 9.1 years; 95% CI 5.2–13.0 years) and had lower education (odds ratio across three education groups, 0.16; 0.06–0.43). However, they did not differ substantially with respect to gender (odds ratio for female gender, 1.60; 0.90–2.88), living alone (odds ratio, 0.67; 0.40–1.11), current smoking at the time of HUNT-2 participation (odds ratio, 0.68; 0.34–1.37), number of non-malignant disorders (odds ratio across four groups, 1.05; 0.80–1.37) or recent GP contact at HUNT-2 participation (odds ratio, 0.71; 0.37–1.36). They were also not significantly different with respect to beyond-local EOD of cancer at diagnosis (odds ratio, 1.15; 0.63–2.09). Proportions with missing data on EOD are summarised for the most common cancer types in the first column of Table 1. Missing EOD was associated with younger age (mean difference, −2.1 years; −3.5 to −0.6) and marginally with female gender (odds ratio, 1.17; 0.98–1.38), higher education (odds ratio across three groups, 1.11; 0.99–1.24) and negatively with living alone (odds ratio, 0.84; 0.70–1.00), but was not associated with smoking (odds ratio, 1.02; 0.86–1.24), number of non-malignant disorders (odds ratio across four groups, 1.04; 0.96–1.14) or recent GP contact (odds ratio, 1.03; 0.82–1.28). There were no associations between case-level anxiety or depression on the HADS and missing information on EOD (odds ratios compared with those with no disorder: 0.73 (95% CI 0.41–1.31) for depression alone, 1.22 (0.65–2.30) for anxiety alone and 1.00 (0.50–2.01) for both anxiety and depression).
Table 1 EOD at diagnosis in new cancer cases within 1–3 years after HUNT-2 participationa
EOD, extent of disease; HUNT-2, second Health Study of Nord-Trøndelag County.
a. All analyses restricted to participants with Hospital Anxiety and Depression Scale data recorded in HUNT-2 and with a first cancer registration 1–3 years later (N = 606, when including participants with unknown EOD).
Characteristics of the 407 participants in the analysed sample are summarised in Table 1 with respect to cancer diagnosis and EOD. Beyond-local EOD of cancer at diagnosis was present in 245 participants (60.2%) and was only significantly associated with current smoking (Table 2). Descriptive data on covariates are summarised for HADS case categories in Table 3.
Table 2 Characteristics of the analysed samplea and associations with beyond-local EOD at cancer diagnosis
EOD, extent of disease; HUNT-2, second Health Study of Nord-Trøndelag County.
a. Restricted to participants with Hospital Anxiety and Depression Scale data present at HUNT-2 and a first cancer registration 1–3 years later.
b. Angina pectoris, asthma, diabetes, epilepsy, hypertension, myocardial infarction, musculoskeletal disorder, respiratory disease, stroke and thyroid disease.
c. N=373 with data on this item.
Table 3 Characteristics of the analysed samplea according to previous anxiety and depression status
EOD, extent of disease; HUNT-2, second Health Study of Nord-Trøndelag County; HADS, Hospital Anxiety and Depression Scale.
a. Restricted to participants with HADS data present at HUNT-2 and a first cancer registration 1–3 years later.
b. Angina pectoris, asthma, diabetes, epilepsy, hypertension, myocardial infarction, musculoskeletal disorder, respiratory disease, stroke and thyroid disease.
c. N = 373 with data on this item.
Of the 407 participants analysed, 306 (75.2%) had no case-level anxiety or depression at HUNT-2 participation, 47 (11.5%) had depression alone, 28 (6.9%) had anxiety alone and 26 (6.4%) had both. Beyond-local EOD at cancer diagnosis was present in 59.8%, 76.6%, 39.3% and 57.7% of these four groups, respectively (Fig. 1). Associations between these groups and EOD were investigated through logistic regression models, the results of which are displayed in Table 4. In summary, adjustment for covariates made little difference to odds ratios for the associations of interest, slightly strengthening those for depression alone and weakening slightly those for anxiety alone. Negative associations with anxiety alone fell below statistical significance in most adjusted models, whereas those for depression alone retained significance in most models; however, confidence intervals for the latter overlapped the null value in the final a priori model, which included cancer type as a covariate. Associations for both anxiety and depression (compared with neither) were close to null values in all models. Further addition of recent GP contact to model 6 in 373 participants with data on this variable gave rise to odds ratios of 2.43 (CI 1.13–5.23) for depression only, 0.50 (0.21–1.20) for anxiety only and 0.86 (0.34–2.15) for anxiety and depression.
Fig. 1 Proportion (%) of sample with beyond-local extent of disease (EOD) at first cancer diagnosis according to anxiety and/or depression caseness 1–3 years previously.
Table 4 Logistic regression analysis of associations between HADS case category and EOD at subsequent cancer diagnosisa
EOD, extent of disease; HADS, Hospital Anxiety and Depression Scale; HUNT-2, second Health Study of Nord-Trøndelag County.
a. Restricting analysis to a lag time of 1–3 years between HUNT-2 and first cancer diagnosis.
b. Entered as ordinal variables, using the categories described in Table 2.
c. Entered as a binary variable (current smoking or not at the time of HUNT-2 participation).
d. As categorised in Table 1.
Discussion
Linking data from a large cross-sectional survey and a national cancer registry, we investigated the association between EOD at first cancer diagnosis and previous symptoms of anxiety and depression. In summary, beyond-local EOD at diagnosis was more likely in participants with previous depressive symptoms and less likely in those with previous anxiety symptoms, supporting our hypothesis. Both associations were marginally stronger in fully adjusted models than in unadjusted, but confidence intervals also widened, and associations were no longer statistically significant according to conventional norms. Where depressive and anxiety symptoms were both present, there was no significant difference in EOD at cancer diagnosis compared with those with no case-level symptoms.
The influence of mental health on mortality and morbidity from other disorders is increasingly recognised,Reference Prince, Patel, Saxena, Maj, Maselko and Phillips1 and there is a growing awareness that the public health effect of mental disorders may be underestimated. Effects of depression on diagnosis and treatment are potential pathways underlying associations with risk of death from cancer. However, associations with EOD at diagnosis have received relatively little research. Prasad et al found that men with depression and an intermediate- or high-risk prostate cancer were less likely to undergo treatment.Reference Prasad, Eggener, Lipsitz, Irwin, Ganz and Hu22 This also appears to be the case where depression is pre-existing before receiving a cancer diagnosis.Reference Oerlemans, van den Akker, Schuurman, Kellen and Buntinx23 Timing of diagnosis is an important factor predicting cancer survival and is determined by both characteristics of the person with cancer (in terms of help-seeking behaviour and time taken to presentation) and characteristics of the healthcare provider (in terms of promptness of investigation, diagnosis and treatment after presentation).Reference Andersen, Cacioppo and Roberts24 In a review of studies of breast cancer specifically, characteristics of the symptoms and personal attributions of these were found to influence delay in presentation, as well as some demographic predictors.Reference Ramirez, Westcombe, Burgess, Sutton, Littlejohns and Richards25 The role of mental disorders in predicting delay in presentation has received relatively little attention. Low trait anxiety was found to be associated with delayed presentation of rectal cancer,Reference Ristvedt and Trinkaus7 particularly in men.Reference Ristvedt and Trinkaus8 In contrast, other studies have found no association between delayed presentation of breast cancer and previous mood disorders, adverse life eventsReference Burgess, Ramirez, Smith and Richards9 or previous psychological treatment.Reference Burgess, Potts, Hamed, Bish, Hunter and Richards11 However, all these studies have relied on retrospective evaluation of mental health, which is likely to be biased in people who have received a cancer diagnosis.
As far as we are aware, only two previous studies have used data on mental health collected before cancer diagnosis. One of these found associations with both previous anxiety/depressive symptoms/disorders recorded on primary care records and longer time from presentation to treatment in people with colorectal cancer.Reference Robertson, Campbell, Smith, Donnan, Sullivan and Duffy12 The other study investigated a small sample women with breast cancer who had previously been screened for mental disorder in a large community survey and found that major depression was associated with a higher risk of late-stage diagnosis, whereas phobia was associated with a lower risk of late-stage diagnosis.Reference Desai, Bruce and Kasl13 In our study, case-level anxiety was, if anything, associated with an earlier stage of cancer presentation, although it was not statistically significant in some of the adjusted model and there was no evidence in any model of a risk association; however, comparability with previous findings is limited because we were not able to examine colorectal cancer specifically with adequate power (unlike the primary care studyReference Robertson, Campbell, Smith, Donnan, Sullivan and Duffy12), and anxiety symptoms measured on the HADS instrument tend toward those of generalised anxiety and panic disorder, rather than phobia (unlike the community surveyReference Desai, Bruce and Kasl13). Case-level depressive symptoms were associated with a higher likelihood of beyond-local EOD to a significant extent in most adjustment models and are thus more consistent with both previous studies, although this could not be demonstrated at statistical significance in the final model, potentially because of insufficient statistical power and/or overfitted covariates. Our conclusion is therefore that it lends support to the study hypothesis, but not to a conclusive extent.
The mechanisms underlying our findings may be complex and multiple. Delay in time of cancer diagnosis (for which EOD was used in our study as a proxy measure) has been divided conceptually into ‘patient’ and ‘provider’ sources, with further subdivision of the former into ‘appraisal’, ‘illness’ and ‘behavioural’ delay, and the latter into ‘scheduling’ and ‘treatment’ delays.Reference Andersen, Cacioppo and Roberts24 Considering ‘patient’ sources, it is possible that depressive syndromes reduce the likelihood or help-seeking (or attendance at a screening programme) for a known early symptom of cancer and that anxiety syndromes increase this. It has been proposed that there is a ‘normal’ response of minimisation during the period of symptom self-appraisal, which may be attenuated in high-anxiety statesReference Ristvedt and Trinkaus7 and could conceivably be exaggerated in depression (or alternatively the normal help-seeking response to a concerning symptom could be inhibited by depressed mood, compounded by accompanying symptoms of poor motivation, guilt, feelings of not being worthy of care and/or nihilism about the future). Consistent with this, a study of presentations to a breast clinic regardless of diagnosis found that higher psychological morbidity was associated with delayed presentation, and that individuals who were least anxious about their symptoms also delayed seeking help.Reference Nosarti, Crayford, Roberts, Elias, McKenzie and David10 It is also possible that an individual's symptom profile may influence physician behaviour; for example, a higher likelihood of referral for further investigation in people who report a given symptom in the context of co-occurring anxiety symptoms and a lower likelihood in those with depressive symptoms. However, our analysis was not designed to investigate these processes in depth and further research is required to identify the most appropriate point of intervention. Of note, an analysis of a large mental healthcare database linked to a cancer registry in south London found no differences in EOD at presentation in people with a range of mental disorders, including depressive and anxiety disorders and those not known to mental healthcare at the time of cancer diagnosis; however, post-diagnosis mortality remained raised in many disorder groups, including depressive disorder, and these associations were independent of EOD at presentation.Reference Chang, Hayes, Broadbent, Hotopf, Davies and Møller26 This suggests that inequalities in care persist after diagnosis, although the extent to which these are accounted for by patient or provider sources remains to be determined. Although no association was found between depressive disorder and EOD at cancer diagnosis in that study, the findings are not necessarily comparable with our own because a diagnosis of depression implies a level of service access that is unlikely to be attained by a community sample screening positive for depressive symptoms.
The negative age-adjusted association between anxiety symptoms in our analysis and EOD at cancer diagnosis is interesting in that the effect, if present, appears to be opposite to that of depression. It is also consistent with previous analyses of mortality data from HUNT-2, where anxiety (identically defined) was found to have a relatively protective effect compared with the risk effect associated with depression.Reference Mykletun, Bjerkeset, Overland, Prince, Dewey and Stewart27 As mentioned, one possible explanation is that people with anxiety who experience early symptoms of cancer are more likely to seek medical help/investigation and/or that GPs are more likely to investigate these symptoms in someone with anxiety. Supporting this, people with case-level anxiety in HUNT-2 reported higher levels of recent GP contact; however, there was no marked association observed between that level of contact and EOD at cancer presentation 1–3 years later. On the other hand, the association was weakened and fell below statistical significance levels following adjustments for other risk factors such as smoking and comorbidity, so at least part of the explanation may lie in healthier lifestyles adopted by people with anxiety, which themselves might correlate with higher likelihood of help-seeking.
Our analysis had the advantage of a very large survey data-set in which anxiety and depressive symptoms were systematically obtained before cancer diagnosis, using a standard and widely applied screening instrument. This rendered the analysis in question at least feasible, although the overlap with cancer registrations was still small. A further advantage of the HADS is that it was specifically developed to ascertain symptoms of anxiety and depression in samples with somatic disorders, focusing primarily on cognitive symptoms. The HADS is therefore less likely than other screening scales (or indeed diagnostic instruments) to misclassify somatic symptoms, an important advantage for the analysis in question. Furthermore, the analysis specifically excluded cancer diagnoses within a year of HUNT-2 participation, reducing the likelihood that HADS caseness was influenced by early cancer manifestations. There is therefore little chance that information bias could explain the results, although we accept in retrospect that the restriction may have been overly conservative and limiting with respect to the sample size. In terms of the analysed sample, both the HUNT-2 survey and CRN are comprehensive and inclusive with respect to the population concerned and we do not believe that selection is likely to have influenced findings apart from non-differential bias owing to missing data. None of covariates had any substantial influence on the associations of interest. However, as with any observational study, residual confounding cannot be ruled out entirely. For example, unmeasured personality characteristics might influence both risk of depression and cancer presentation, as might a family history of either/both mental disorders and cancer.
The principal limitation of the study was in the size of the analysed sample, which did not allow associations with specific cancers to be investigated and in which groups defined by the presence of case-level anxiety and depression were relatively small, despite the large size of the survey sample followed in this respect. Non-significant associations should therefore be treated with caution because of lack of statistical power. Furthermore, there was a high proportion of missing information about EOD in the national register, although we found no evidence that missing data were differential by gender or exposure status, only that those with missing staging were older (P < 0.001); no attempts were made at multiple imputation because of the likelihood that missing data were not at random. Considering measurement accuracy, it should be borne in mind that the HADS is a brief screening instrument designed for ascertaining symptoms rather than disorders, and it was only administered on a single occasion, which may have underestimated the associations of interest. Finally, EOD can only be considered as a proxy measure of timing of diagnosis, and factors associated with EOD cannot be assumed to be causing a delay in diagnosis.
In conclusion, in people who develop cancer, we investigated previously recorded depressive and anxiety symptoms as risk factors for beyond-local EOD at diagnosis. In analyses adjusted for age alone, depressive symptoms were associated with higher likelihood of beyond-local EOD and anxiety symptoms with a lower likelihood of this outcome. These parallel associations of depressive and anxiety symptoms with cancer mortality previously reported in the same cohort (i.e. higher risk associated with depressive symptoms, lower risk with anxiety symptoms).Reference Mykletun, Bjerkeset, Dewey, Prince, Overland and Stewart4 However, importantly, neither association with EOD was statistically significant in fully adjusted models. For anxiety, this might be explained by inclusion of health and lifestyle factors, such as smoking, as covariates so that better health, and therefore possibly increased help-seeking, might have accounted for the protective association. For depression, coefficients remained relatively unaltered and close to statistical significance in most models, so an independent association might have been observed in a larger sample. Considering implications, at the very least the findings suggest that associations with mental health conditions are potentially complex; in this instance, suggesting that two very closely related syndromes (depression and anxiety) may have opposing influences on help-seeking behaviour and/or other factors determining cancer detection. Similar models of investigation could be informative with respect to other disorders, such as cardiovascular disease, whose prognosis is also known to be associated with mental health status and whose presentation and subsequent management have important implications on outcome. Physical health complaints, perhaps particularly in depression, risk being overlooked and attributed to the mental disorder, so there are important potential implications for primary and secondary care vigilance. The potential influence of mental disorder symptoms on how important somatic disorders present to healthcare (and/or are detected and diagnosed) clearly needs to be borne in mind when considering ways in which well-recognised health inequalities in people with mental disorders are to be addressed.
Funding
No specific funding was obtained for this analysis and there was no study sponsor. R.S. and M.H. are part-funded by the National Institute for Health Research Biomedical Research Centre at South London and Maudsley NHS Foundation Trust and King's College London. R.S. has received research support in the past from Janssen, GSK, Roche and Takeda. A.M. received funding from Research Council of Norway at the time of the study and later.
eLetters
No eLetters have been published for this article.