Screening for cancer-related distress: When is implementation successful and when is it unsuccessful?

Abstract

Objective. Screening for distress is controversial with many advocates and detractors. Previously it was reasonable to assert that there was a lack of evidence but this position is no longer tenable. The question is now: what does the evidence show and, in particular, when is screening successful and when is screening unsuccessful? The aim of this paper is to review the most up-to-date recent findings from randomized and non-randomized trials regarding the merits of screening for distress in cancer settings. Methods. A search was made of the Embase/Medline and Web of knowledge abstract databases from inception to December 2012. Online theses and experts were contacted. Inclusion criteria were interventional (randomized and non-randomized) trials concerning screening for psychological distress and related disorders. Studies screening for quality of life were included. Results. Twenty-four valid interventional studies of distress/QoL screening were identified, 14 being randomized controlled trials (RCTs). Six of 14 screening RCTs reported benefits on patient well-being and an additional three showed benefits on secondary outcomes such as communication between clinicians and patients. Five randomized screening trials failed to show any benefits. Only two of 10 non-randomized sequential cohort screening studies reported benefits on patient well-being but an additional six showed secondary benefits on quality of care (such as receipt of psychosocial referral). Two non-randomized screening trials failed to show benefits. Of 24 studies, there were 17 that reported some significant benefits of screening on primary or secondary outcomes, six that reported no effect and one that reported a non-significantly deleterious effect upon communication. Across all studies, barriers to screening success were significant. The most significant barrier was receipt of appropriate aftercare. The proportion of cancer patients who received psychosocial care after a positive distress screen was only one in three. Screening was more effective when it was linked with mandatory intervention or referral. Conclusions. Screening for distress/QoL is likely to benefit communication and referral for psychosocial help. Screening for distress has the potential to influence patient well-being but only if barriers are addressed. Quality of care barriers often act as a rate limiting step. Key barriers are lack of training and support, low acceptability and failure to link treatment to the screening results.

Distress is the experience of significant emotional upset arising from various physical and psychiatric conditions [1]. Screening for distress is relatively new compared with screening for depression which has been more extensively investigated in a variety of settings. However, screening for distress is controversial. The evaluation of evidence regarding screening for distress should be no different to the evaluation of any other screening target such as screening for prostate cancer or cervical cancer. Several authors have put forward a coherent case against routine screening. These views are importance because screening is not so overwhelmingly effective and not without cost, such that no scrutiny of the evidence is needed. A considered negative view actually helps us decide how can we be sure if screening works? Also if screening is only partially successful, can improvements be made such that adoption into routine care makes clinical and financial sense? Screening has been suggested to improve patient outcomes in depression presenting in primary care, but positive benefits have equally been disputed [2,4,5]. The same argument for and against screening has played out in cardiovascular settings [3,5]. Fortunately, we have the opportunity to learn lessons from an extensive literature concerning screening for depression in primary care and other medical areas [4]. One lesson is that when the results of individual studies are mixed then it is difficult for reviewers to avoid confirmatory bias when evaluating the evidence. This particularly applies to non-meta-analyses, although no method is entirely exempt from the possibility of bias. This has been very well-described from the perspective of screening for depression in primary care when two thorough reviews came to entirely opposite conclusions [5].

When evaluating screening for distress, the ideal comparison is with treatment as usual. Yet treatment as usual is by no means uniform. Treatment as usual may be high or low quality, high or low resource. It is very likely that routine screening would fail to show benefits when compared to an unscreened cohort seen by expert/interested clinicians who reliably offered a wide choice of patient friendly resources. However, this scenario is not common and almost all major centers show considerable variability in psychosocial care [6]. The introduction of screening reduces that variability at the point of diagnosis, but if treatment is not offered then screening is fruitless. For this reason, the challenge to centers screening for distress is to ensure effective treatment follows accurate diagnosis. When we evaluate screening studies, we are most interested in added value, that is, the additional merit of screening that would not otherwise be achieved by routine clinical judgement. Although routine clinical judgement is notoriously inaccurate compared with our current gold standards (e.g. DSMIV diagnoses) some cases are picked up and many people without distress are identified. Most physicians working with cancer patients are not confident in dealing with distress, most do not use any screening instruments and most have little education and training in psychosocial issues [7]. Figures from our Leicester cancer center suggest frontline clinicians have about 50% sensitivity and 80% specificity when looking for distress [8]. About half of identified cases are offered timely, appropriate treatment. Results are broadly consistent with other centers which also find approximately 20–30% of people with unmet psychosocial needs will have already been recognized and treated at any one point in time [9]. The purpose of screening is to improve on this figure, to address unrecognized problems in the remaining 70–80%. In short, screening aims to reduce inequalities in diagnosis that result from differing clinician abilities. In a well-designed randomized controlled trial (RCT) of screening versus clinical judgement (diagnosis), it would be reasonable to test the yield of screening versus judgement for cases not previously identified, providing this standard is applied equally to both arms. Yet, it is also reasonable to test the yield of screening versus judgement for all cases (whether or not previously identified) providing the screening study clarifies how many identified patients desire psychosocial help or referral because the fundamental aim of psychosocial care is to provide timely, appropriate and acceptable care for patients with current self-reported unmet needs regardless of their cancer stage, cancer diagnosis or past treatment history.

Should the target of screening be distress?

Screening must have a worthwhile treatable target and there has been a dispute whether distress is really a disabling condition. In recent years several organizations have promoted distress, rather than depression, as the key emotional patient-reported outcome measure in cancer care [10]. The distress concept has the advantage of lower perceived stigma than depression, and broad acceptability to patients. Its main disadvantage is that distress is poorly operationalized, and it corresponds only approximately to known psychiatric disorders. Distress can be mild but when moderate or severe can be considered a generic category of emotional suffering that encompasses psychiatric conditions such as depression, anxiety, and adjustment disorder in addition to non-psychiatric psychological and practical concerns [11]. Distress is not a specific category in Diagnostic and statistical manual of mental disorders, 4th ed. (DSMIV) or International classification of diseases, 10th ed. (ICD10) and therefore should not be considered a medical condition per se but a symptom. Yet there is accumulating evidence suggesting that the presence of distress is associated with reduced health-related quality of life [12], poor satisfaction with medical care [13] and possibly reduced survival after cancer [14]. A medical analogy is that screening for distress is like screening for high glucose, whereas identifying depression is analogous to detecting diabetes. Diabetes mellitus is only one cause of hyperglycemia, but hyperglycemia is a significant problem on its own. Distress, unmet needs and related psychiatric disorders are certainly treatable conditions [15]. Distress is closely linked with unmet needs and it is well-documented that many cancer patients report that their psychosocial and physical needs are not met [16].

National Screening Guidelines

Details of how to screen and how often to screen are subject to much local variation and few countries have any unified national policy [17,18]. Guidelines have not been sufficiently evidence-based to make a case that convinces both advocators and detractors of screening. Those against routine screening raise several worthwhile cautions. First, that screening should apply only to those not already currently recognized as depressed in receipt of treatment. Second, that those who screen positive often do not accept the treatment that is offered [19]. Third, the same treatment and care resources should be available to both groups (screened and not-screened) to effectively isolate the effect of screening per se. Fourth, screening routinely may be inefficient given that many people have very mild complications. Fifth, screening can be resource intensive and can be a burden to staff and patients. These arguments should be considered whilst reviewing the forthcoming evidence below.

Evaluation of distress screening studies

Implementation can be defined as the ‘systematic introduction of innovations and/or changes of proven value, the aim being that these are given a structural place in professional practice, in the functioning of organizations or in the health care structure’ [20]. Screening implementation is the process whereby a screening method is applied to clinical practice, ideally under scrutiny in order to clarify hazards and benefits. Phases in the development and testing of a screening tool have been reported [21]. Several groups have reviewed diagnostic validity studies in depth but most have concentrated on depression per se [22–24] and meta-analyses have been carried out on both depression tools [25] and on distress tools [26]. Before discussing implementation studies it is essential to briefly review the methodology underlying screening studies (Table I) [27]. Once a screening tool has been developed and tested for potential accuracy against an accepted gold standard, it can be evaluated in a clinical setting. This is the implementation phase. The implementation can be non-comparative, or observational. Such studies are not without value. For example, the effect of screening on quality of care (process measures) or patient reported outcomes can be monitored using current or historical data. Observational studies will reveal how well screening is working, but will not reveal how much better screening is over usual care. For this, an interventional screening study is required. These can be randomized or non-randomized. In the typical randomized study, two equal groups of clinicians, or in the case of ‘cluster randomization’, two centers, are randomized to have either access to screening versus no access to screening. A variant on this design is to randomize two groups to have either access to results of screening or screening, but no feedback of the results of screening. In effect it is feedback of results that are randomized not screening. Theoretically this may help distinguish which effects are related to application of the screener and which to the receipt of screening results.

Table I. Methodology of screening studies.

Type screening study	Purpose of study	Description of study
Diagnostic validity study	Establish diagnostic accuracy of a tool against a gold standard instrument	A screening tool is tested against a criterion (gold standard) in a real world sample generating the sensitivity and specificity of the tool, as well as positive and negative predictive value which depend on the cut-off chosen and the prevalence of distress.
Non-randomized sequential cohort Implementation study	Establish the added value of screening on patient outcomes and quality of care	The screening tool is evaluated clinically in one group of clinicians with access to screening (or results of screening) compared to a second group (typically a historical group or second centre) who do not access to screening (or results of screening)
Randomized controlled Implementation study (screen vs no-screen)	Establish the added value of screening on patient outcomes and quality of care, controlling for baseline variability	Two equal groups of clinicians (or in the case of ‘cluster randomization’ centres) are randomized to have either access to a screening method vs no access to screening.
Randomized controlled Implementation study (screen + feedback vs screen no feedback)	Establish the added value of screening feedback on patient outcomes and quality of care, controlling for baseline variability	Two equal groups of clinicians (or in the case of ‘cluster randomization’ centres) are randomized to have either access to results of screening vs screening but no access to the results of screening (screen no feedback).
Observational Implementation screening study	Establish effect of screening on clinical practice (uncontrolled)	A screening tool is introduced in clinical practice and the effect on quality of care (process measures) or patient reported outcomes monitored. This can be conducted using current or historical screening data

The next methodological question is what outcome is relevant to screening studies? Historically the main outcome of interest has been patient well-being (also known as patient reported outcomes measures or PROMS). This review will focus on this key outcome but readers should be aware of secondary outcomes that are of interest but beyond the scope of this review. Secondary outcomes of interest are clinician behavior/quality of care. Clinician behavior includes the number of accurate diagnoses recorded, doctor-patient communication, referrals made to specialist services and psychosocial help given by clinicians. These ‘quality of care’ markers are sometimes called process measures but can influence PROMs. For example, Carlson et al. (2010) found that the best predictor of decreased anxiety and depression was receipt of referral to psychosocial services [28]. If screening studies show benefits in quality of care or clinician behavior but not patient well-being, then this suggests there are significant barriers to care downstream of the screening process. An important measure in all studies is acceptability of the screening program to patients and clinicians. This can be measured by satisfaction scores or by proxy measures such as uptake and participation.

The aim of this paper is therefore to review the latest evidence concerning the evidence for and against screening for distress/QoL and summarize the lessons from randomized studies and non-randomized studies which have been successful and unsuccessful in terms of primary (and to a lesser extent secondary outcomes and acceptability).

Methods

A search was made of the Embase/Medline and Web of knowledge abstract databases. Detailed methods are as described in a previous study, but updated to December 2012 [30]. The inclusion criteria were randomized and non-randomized interventional implementation studies regarding the effects of distress screening on key outcomes. All potentially valuable studies were included regardless of their outcome. The key outcomes were change in patient well-being, reported acceptability, receipt of psychosocial treatment (or referral for treatment) and clinician communication. Previous reviews were searched as well as theses and experts contacted [24,29,30]. We examined the following methodological aspects of each study: design and methods, setting and sample, uptake, predictors and confounders. Results were stratified into successful and unsuccessful screening studies based on the findings of at least one statistically significant (p-value of 0.05 or lower) positive primary or secondary outcome (hereby defined as a positive trial) a non-significant effect or a deleterious effect (hereby defined as a negative trial).

Results

From a total of 520 studies retrieved from the literature searches, 14 randomized trials were identified regarding the effect of screening for psychological distress and a synopsis is shown in Table II. A further 10 non-randomized studies were identified that measured changes in distress or related outcomes before and after screening without randomization. Several other studies with psychological PROMs were not included as they did not randomize or evaluate the effect of screening itself.

Table II. Brief summary of successful and unsuccessful distress screening implementation studies.

Author/Country	Screening target	Screening beneficial?	PROs improved?	Referrals improved?	Communication improved?	Acceptability of screening?
Randomized Unsuccessful
Maunsell et al. (1996) [35] Canada	Distress Depression	No	No	NR	No	NR
Rosenbloom et al. (2007) [36] USA	Mood Quality of life	No	No	NR	NR	NR
Mills et al. (2009) [42] UK	Quality of life	No (deleterious)	No	NR	Yes but not significantly	High
Braeken et al. (2011) [37] Germany	Distress	No	No	Yes (but not significantly)	NR	Mixed
Hollingworth et al. (2012) [38] UK	Distress Quality of life	No	No	NR	NR	High
Randomized successful
Sarna (1998) [31] USA	Distress	Yes	Yes	NR	Yes	NR
McLachlan et al. (2001) [32] Australia	Distress Depression Unmet needs	Partial (in depressed patients)	Yes (in depressed only)	NR	Yes (in depressed only)	NR
Detmar et al. (2002) [39] Netherlands	Quality of life	Partial	No	NR	Yes (but only for social functioning, fatigue and dyspnea)	NR
Velikova et al. (2004) [33] UK	Distress Quality of life	Yes	Yes	NR	Yes but not significantly	Mixed
Macvean et al. (2007) [43] Australia	Quality of life Unmet needs Depression	Yes	Yes (depression)	NR	NR	Good
Girgis et al. (2009) [44] Australia	Unmet needs Depression Quality of life	Partial (in communication and action)	No	Yes and significant (but full data not presented)	Yes and significant (but full data not presented)	High
Carlson et al. (2010) [28] Canada	Distress	Yes (in breast and lung cancer)	No	Yes significantly	NR	High
Carlson et al. (2012) [45] Canada	Distress	No	Yes	Yes (access to services) significantly	NR	High
Klinkhammer-Schalke et al. (2012) [34] Germany	Quality of life	Yes	Yes	NR	NR	Not reported
Non-randomized unsuccessful
Boyes et al. (2006) [46] Australia	Depression/anxiety Unmet needs	No	No	NR	NR	Yes
Mitchell et al. (2012) [47] UK	Distress Depression/anxiety/anger	Not significantly	No	NR	NR	Partial
Non-randomized successful
Taenzer et al. (2000) [48] Canada	Quality of life	Partial (in communication and action)	Yes	NR	Yes but not significantly	NR
Pruyn et al. (2004) [49] Netherlands	Distress	Yes	NR	Yes significantly	Yes but not significantly	Duration of consultations decreased & screening acceptable to 77% of patients
Bramsen et al. (2008) [50] Netherlands	Distress Quality of life	Yes	Partial	Yes significantly	NR	NR
Hilarius et al. (2008) [51] Netherlands	Quality of life	Partial (in recognition and action)	No	Yes (but not significantly)	Yes significantly overall	NR
Thewes et al. (2009) [52] Australia	Distress Unmet needs	Partial (in referral delay)	No	Yes but not significantly	NR	Yes
Shimizu et al. (2010) [53] Japan	Distress	Partial (in referral)	No/Unknown	Yes significantly	NR	NR
Ito et al. (2011) [54] Japan	Distress	Partial (in referral delay)	No	Yes, significantly	NR	NR
Grassi et al. (2011) [55] Italy	Distress	Partial (in referral)	No	Yes, significantly	No	NR

NR, not recorded.

Brief summary of successful and unsuccessful distress screening implementation studies

Summary of evidence.

Twenty-four valid interventional studies of distress/QoL screening were identified, incorporating 14 RCTs and 10 sequential cohort studies. Although patient well-being often improved, it did not necessarily show differential improvement compared with the control arm. Only six of 14 screening RCTs reported added benefits on patient well-being. An additional three showed benefits on secondary outcomes such as communication between clinicians and patients. Five randomized screening trials failed to show any benefits. Similarly, although two of 10 non-randomized sequential cohort screening studies reported benefits on patient well-being, an additional six showed secondary benefits on quality of care (such as receipt of psychosocial referral). Only two non-randomized screening trials failed to show any significant benefits.

Thus an appraisal of 24 screening implementation studies shows that there were 17 studies that reported some significant benefits of screening on primary or secondary outcomes and six that reported no significant effects and one that reported a non-significantly deleterious effect upon communication. The principal secondary benefits appear to be on referral to specialist services and communication. Distress and QoL screening appear to open the door to a dialogue with clinicians who can then determine which unmet needs have contributed to distress. As such distress screening can probably be supplemented by an unmet needs checklist (such as the NCCN's problem list). Acceptability was only studied in depth in 12 out of 24 studies. Of these, acceptability was good to very good in nine studies but mixed in three studies, but never poor. Overall then, the acceptability of distress/QoL screening appears to be satisfactory. At the study level additional lessons are apparent (below).

Lessons from successful randomized screening studies.

Sarna (1998) conducted a small randomized trial in 48 patients whereby the results of screening with the Symptom Distress Scale (SDS), Hospital Anxiety and Depression Scale (HADS) and Karnofsky Performance Status (KPS) were fed back or not fed back to clinical nurses according to randomization [31]. Over six months of follow-up, symptom distress in the feedback group declined, but in the no feedback group it increased and the difference was statistically significant by six months. McLachlan et al. (2001)'s RCT involving quality of life, depression and unmet needs was the first well-powered study (450 patients) [32]. Patients completed self-reported questionnaires via a touch-screen computer with results feedback to the doctor and formulation of an individualized management plan in those with positive screens. In those depressed at baseline, there was a significantly greater reduction in depression for the intervention arm, indicating that screening/interventions most benefit those with most distress at baseline and that screening with resources is likely to be more effective than screening alone. Velikova and colleagues (2004) recruited 28 oncologists treating 286 cancer patients and randomly assigned them to screening along with feedback or screening alone (called attention-control) or a no screening condition using EORTC QLQ-C30 and touch-screen HADS [33]. A positive effect on emotional well-being was seen in the intervention with feedback versus control group suggesting screening with feedback is the most effective option. Acceptability, however was modest. Carlson et al. (2010) [28] took the Velikova et al. model and included minimal screening (no feedback), full screening (with feedback) [33] but added screening with feedback and optional triage and referral (enhanced screening). In breast cancer patients the full screening and triage groups both had lower distress at follow-up compared with minimal screening. Recently, Klinkhammer-Schalke for the Regensburg QoL Study Group (2012) randomized 200 breast cancer patients to receive either feedback of low QoL (with a report sent to clinician), or standard care [34]. Outcome QoL favored screening suggesting perhaps feedback of only the significant results are needed during screening.

Lessons from unsuccessful randomized screening studies.

Maunsell et al. (1996) conducted an RCT of telephone GHQ-20 screening every 28 days (n = 123) against basic psychosocial care only (n = 127) and screening incorporated an automatic referral process [35]. However, distress decreased over time in both groups with little to differentiate between groups and no additional benefit of screening hinting at high quality care in the control arm. Rosenbloom et al. (2007) randomly assigned 213 metastatic patients to feedback or no feedback of Functional Assessment of Cancer Therapy- General (FACT-G) results [36]. No effect of PROMs was found. Mills et al. (2009) also found null results using a focussed QoL diary completed at home. Braeken et al. (2011) conducted an innovative study using radiotherapists who were asked to apply a 24-item Screening Inventory of Psychosocial Problems (SIPP) but found no significant benefit attributable to screening, perhaps because the burden fell to busy frontline clinicians who had difficulty with implementation [37]. Similarly, Hollingworth et al. (2012) did not find significant differences in Profile of Mood States (POMS) or quality of life when screening was completed by frontline radiographer/nurses using the (DT) and problem list [38]. From these results it appears that frontline clinicians struggle to adapt screening into routine care.

Discussion

Screening for distress in cancer is a rapidly evolving field with an appreciable body of evidence. Previous work has largely focussed on the development and diagnostic validity testing of tools for measuring cancer-related distress. Despite strong recommendations of many professional societies and accreditation agencies, valid cautions against premature adoption of screening exist. Previously, it was reasonable to assert that there was a lack of evidence regarding distress screening but with 24 implementation studies this position is no longer tenable with one exception: screening in advanced cancer and palliative settings. Only three implementation studies have examined screening patients with advanced cancer with mixed results [31,39,40]. Overall, results of 24 screening implementation studies show that there are 17 studies reporting some statistically significant benefits of screening on primary or secondary outcomes. For those (apriori) advocates of screening this may be disappointing as six of 14 screening RCTs reported added benefits on patient well-being. For those (apriori) detractors of screening these findings may also be surprising, 17 of 24 implementation studies did reveal some benefit (over and above usual care) albeit often involving secondary outcomes, such as referral to specialists or communication.

How does this evidence inform the cautions against screening mentioned in the introduction? The first caution is that screening should apply only to those not already currently recognized as depressed/distressed and in receipt of treatment [19]. Although this has rarely been addressed Braeken et al. (2011) found that of those who received a referral in the screening RCT, 22% of referred screened patients were previously identified, and 29% of non-screened referred patients were previously identified [37]. In other words the yield was reduced in both screened and non-screened arms by taking into account previous care. The second caution is that those who screen positive often do not accept the treatment that is offered [19]. This is a genuine barrier to receipt of care. Carlson et al. (2012) found that over 12-months follow-up after screening, 20% received services in the screen and triage arm compared with 15% in the screen alone arm [28]. The third caution is the same treatment and care resources should be available to both groups (screened and not-screened) to effectively isolate the effect of screening. In fact, this has been extensively studied in the feedback implementation studies which compare screening with versus without feedback of results. In both arms care is typically treatment-as-usual. From eight feedback versus no-feedback implementation studies, six have found superiority of screening in relation to primary or secondary outcomes, and two have found no effect. The fourth caution is that screening routinely may be inefficient given that many people have very mild complications. Both screening and clinical judgement are more accurate when focussing on more severe cases, however the majority of burden resides in those with mild and moderate disease. The fifth caution is that screening can be resource intensive and can be a burden to staff and patients. This caution is partially upheld, whilst acceptability of screening is generally good, when conducted by frontline clinicians it is often perceived as burdensome. This is somewhat alleviated when screening is brief, has tangible benefits, associated with resources and staff support or when it is conducted in the waiting room screening or using computerized touch screens.

Across all studies, barriers to screening success were significant. At the clinician level the main barriers to screening are lack of time, lack of training and low personal skills or confidence. At the organizational level, barriers include lack of resources and the absence of a screening strategy [7]. However, from this research, the main barrier to successful implementation appears to be receipt of appropriate aftercare. The proportion of cancer patients who received psychosocial care after a positive distress screen was only 20–30%. This shows that aftercare is probably the key rate-limiting step. Screening was more effective when screening was linked with mandatory intervention or referral. This should take the form of a distress management plan to ensure that clinicians systematically act on screening results, and to ensure the healthcare system has resources for helping clinicians manage distress. A positive screening should be followed by thorough clinical assessment and competent management [41]. Depending on the needs identified for specific populations, the actions that follow screening could involve, e.g. a stepped approach, ranging from group-based psycho-education for people with mild–moderate distress to structured individual therapy for those with high distress.

This analysis of the randomized trials and non-randomized implementation studies suggests that some caution regarding systematic routine screening is rational but that evidence does show that screening for distress/QoL has modest but significant benefits largely on quality of care. Additional unmeasured benefits may include feedback on the prevalence of distress to healthcare providers that can be used to directly help patients but also to improve the service delivery system. Audit of systematic assessment is mandatory for service improvement, and a very short step to screening itself. Factors that can influence the success of screening are becoming clearer. It does no longer seems tenable to screen only for one or two psychiatric disorders (such as depression, anxiety), worthy though these target are. Multi-domain screening incorporating unmet needs is much more likely to benefit patient well-being as a whole. Without addressing aftercare, systematic adoption of distress screening in clinical practice is probably not worthwhile. By addressing aftercare, systematic adoption of distress screening in clinical practice is probably worthwhile but issues of acceptability, resources and clinician support must not be overlooked. Key barriers that prevent screening being effective appear to be the same barriers that prevent high quality of psychosocial care in general. Namely, availability and acceptability of a range of suitable treatments, availability and acceptability of experts (e.g. psychologists, psychiatrists) in psychosocial care. In short, screening success may be determined by two key factors: acceptability and resources.

Acknowledgements

Thanks to Amy Waller and Linda E. Carlson who helped with extraction and interpretation of several studies discussed. Thanks also to Christine Clifford for additional advice.

Declaration of interest: The author report no conflicts of interest. The author alone is responsible for the content and writing of the paper.