Audit of prostate cancer

Public Health (1999) 113, 157±160 ß R.I.P.H.H. 1999 http://www.stockton-press.co.uk/ph

Audit of prostate cancer: validity and feasibility of registry-based staging P Silcocks1,2*, P Needham2 and F Hemsley2 Trent Institute for Health Services Research, University of Nottingham, Queen's Medical Centre, Nottingham; and 2Trent Cancer Registry, Weston Park Hospital NHS Trust, Shef®eld, UK

1

The sample consisted of 251 prostate cancer patients who were newly registered during 1994 from a Regional Cancer Registry. It was possible for 96% of cases to be staged by Registry clerical staff, but none had been formally staged by clinicians. In 45% of cases tumour extent was based only on Digital Rectal Examination (DRE), and on 32% there was no assessment found. 81% had no investigation of nodal status and 16% of cases had echography other than trans-rectal ultrasound. To ®nd metastases 18% had scintigraphy and 13% had a plain chest X Ray while 61% had no investigation. Cox regression gave age-adjusted hazard ratios relative to stage A (95% CI) of 2.94 (1.59 to 5.44) for Stage B, 6.43 (3.37 to 12.24) for Stage C, and 9.03 (5.12 to 15.95) for stage D. Key lessons were:  Registry clerical staff can abstract stage-related information from a high proportion of case-notes even if stage is not explicitly recorded by clinicians;  The validity of registry based staging information is supported by survival analysis;  This activity has resource implications and would be more appropriately performed by a member of the clinical team. Readers are referred to the companion paper by the authors of this paper, P Silcocks, P Needham, F Hemsley, titled Audit of prostate cancer: lessons learnt for current clinical practice, surrogates for quality of care and standardisation and quality assurance, which covers the clinical=technical aspects and is published in Public Health 1999 113 161 ± 164. (This issue). Keywords: cancer registration; stage; Cox regression; audit

Introduction Following the Calman-Hine recommendations for the reorganisation of cancer services,1 emphasis has been placed on identifying outcome measures for their evaluation and monitoring. Survival data (an obvious measure of outcome) is readily available from cancer registries, but comparisons need to be corrected for case-mix at diagnosis, often expressed as tumour stage (an index of the extent to which a tumour has spread). Besides indicating prognosis, other reasons why stage should be recorded2 are to:    

Aid the clinician to plan treatment; Help evaluate the results of treatment; Facilitate information exchange among treatment centres; Contribute to human cancer research in general.

clinical coders extracting the relevant information from discharge summaries or medical notes. Treatment details received by TCR are necessarily incomplete as these are based only on the information available at the time the original coding is undertaken. As Trusts are separate organisations the staging data they supply may not be consistent with each other. Within Trusts coding may be more or less consistent depending on whether there is a high proportion of cancer patients and specialisation of the work of cancer coding. If registry staff abstracted staging information (in the same way as death certi®cates are coded centrally at the Of®ce for National Statistics) consistency might be greater. This would also give greater ¯exibility to change staging rules as a smaller number of staff would need to be retrained. In order to develop the TCR services and data set on these lines we wanted to:

At present Regional Cancer Registries are only required to record tumour stage for breast, cervical and colo-rectal cancers and malignant melanomas. In some registries speci®cally trained registry clerical staff (Tumour Registrars, in American terminology) routinely collect stage on all solid tumours by active data collection through examination of medical notes Ð either as part of registration or an `add-on' activity. At Trent Cancer Registry (TCR) registration is principally electronic. Information abstracted from the Patient Administration System (PAS) is forwarded on tapes from each Trust to the Registry. This relies on each Trust's

 assess if cancer coding data could be collected by Registry clerical staff without previous experience and its feasibility for clinical coders;  estimate cost of collecting such information;  test the validity of the stage information collected (by survival rates).

*Correspondence: Dr Paul Silcocks, Trent Institute for Health Services Research, University of Nottingham, Queen's Medical Centre, Nottingham, NG7 2UH, UK. Accepted 22 April 1999

Choice of subjects

The exercise was included in a routine audit of data supplied to the Registry. Prostate cancer was chosen because it is stageable, future screening trials are possible and one of the authors (PS) has a personal interest in the subject. Subjects and methods Three-hundred new registrations were chosen by a simple random sample of the Registry database (from all Districts

Prostate survival P Silcocks et al

158

in the Trent Region for 1994). This sample size was determined by workload rather than statistical considerations. (It had to be included as part of a routine 2% audit of the data supplied to TCR).

performed at TCR by a statistical trainee employed by the Registry.

Staging

Comparison of survival was performed by Cox regression4 using the packages KMSURV5 and COXSURV6 to plot

The broad principles of the Tumour Nodes Metastases (TNM) system were adopted. Our original intention was to record TNM stage if given by the consultant but if not, to impute TNM stage. However the 1992 revision of TNM introduced new categories (T1c, T2c and T3c) but did not indicate whether the pro®les for each TNM stage should be revised. In practice the detailed data needed to categorise, for example ` . . . incidental ®nding in 5% or less of tissue', ` . . . tumour involves more than half a lobe or less . . . ', `or size and number of lymph nodes . . . ', were simply not available in case notes. We therefore collected data on the three axes of T, N and M, using the broad T1 T2 T3 and T4 categories and coding nodal involvement and metastases as NX N0 or N1, MX M0 or M1, as far as we could. This is in line with TNM rules which require the lower category to be coded if there is doubt. We then converted these broad TNM categories to stage as de®ned by the Whitmore system3 (Figure 1), counting the `X' (not assessed) category as if it was negative. Certain rules were adopted to impute T category from written descriptions. For example, given that T1 tumours are impalpable, a clinical description of `smooth, enlarged' counted as T1. Terms such as `hard, irregular', `hard and craggy' counted as T2, but if there was doubt between two categories the TNM rule of assigning the lower one was adopted. For T3 we required either description of spread through the capsule or uncertainty that it was T4 (needing description of ®xation). In addition we accepted abdominal ultrasound as a test for para-aortic nodes. For each of the T, N, and M axes we also recorded the basis of the categorisation, for example histology, CT scan. Lastly, investigations were accepted as staging-related if they were performed in the period two weeks before to two months after the anniversary date (loosely, the date of diagnosis).

Statistical methods

survival and estimate regression coef®cients respectively. Other analyses were performed using EXCEL.7

Results The selection of the study sample is outlined in Figure 2. After excluding four duplicate registrations 251 case notes were available for study. One Trust could not provide any patient notes. Stage depends on separate assessment of T, N and M axes (Table 1). In 45% of cases the tumour extent was based only on digital rectal examination (DRE), though for 32% we could not ®nd any assessment. Nodal involvement was rarely investigated formally: 81% had no investigation and 16% of cases had ultrasound, but not trans-rectal ultrasound (TRUS). For metastases 61% had no investigation, 18% had scintigraphy and 13% had merely a plain chest X Ray. In virtually all instances the clinician did not record an explicit N or M category (97% and 96% respectively), but a T category was recorded by the clinician in 17% of subjects (Table 2). As imputed by us, over 97% were stageable, with similar proportions in each category (Table 2). Evaluation of the effect of stage on survival by Cox regression (with truncation at 31=07=97) showed distinct

Procedures A team of four Registry staff undertaking audit work, including the Acting Quality Assurance Manager, were trained, using case-notes from a large teaching hospital for illustration, by the Registry's Medical Advisor (PS). The team was accompanied by PS for initial visits but for later visits, the team split into pairs for the smaller hospitals. Queries were resolved by PS, and data entry was

Figure 1 Staging categories.

Figure 2 Origin of study sample.

Prostate survival P Silcocks et al

159

Table 1 Investigations done to establish stage of prostate cancer by tumour, nodes and metastases (TNM) category 95% CI)

Table 3 Percentage of prostate cancer cases in each stage (95% CI)

Tumour Rectal examination only (DRE) No investigation recorded DRE plus other (TRUS=CT, US) US only Other=NOS Nodes No investigation recorded Non-TRUS U=S CT scan Histology Metastases No investigation recorded Scintigraphy alone Plain XR alone MRI scan Other & Combined investigations N ˆ 251

A B C D Inadequate data to stage N ˆ 251

45% (39.1% ± 51.8%) 32% (26.5% ± 38.4%) 7% (4.0% ± 10.6%) 6% (3.4% ± 9.7%) 10% (6.2% ± 13.9%) 81% (75.9% ± 85.9%) 16% (11.6% ± 21.1%) 2% (0.6% ± 4.6%) 1% (0.1% ± 4.6%) 61% (54.6% ± 67.0%) 18% (13.4% ± 23.2%) 13% (8.9% ± 17.5%) 0.4% (0.0% ± 2.2%) 8% (4.9% ± 12.0%)

TRUS: Trans-rectal Ultrasound; MRI: Magnetic Resonance Imaging; CT: Computerised Tomography; DRE: Digital Rectal Examination; US: Ultrasound; NOS: Not otherwise speci®ed.

Table 2 Proportion of prostate cancer case notes giving tumour, nodes and metastases (TNM) categories (95% CI) Classi®ed by clinician

Tumour Nodes Metastases N ˆ 251

Yes

No

Missing from record sheet

17% 12% ± 22.4%) 1% (0.4% ± 2.8%) 2% (0.6% ± 4.6%)

80% (74.2% ± 84.5%) 97% (94.3% ± 98.9%) 96% (92.8% ± 98.1%)

3% (1.4% ± 6.2%) 2% (0.6% ± 4.6%) 2% (0.6% ± 4.6%)

35% (29.5% ± 41.7%) 24% (18.8% ± 29.7%) 12% (8.5% ± 17.1%) 25% (19.9% ± 30.9%) 3% (1.4% ± 6.2%)

and highly signi®cant survival curves for each stage (Figure 3). This was maintained even when age was added to the model, with age-adjusted hazard ratios relative to stage A (95% CI) of 2.94 (1.59 to 5.44) for Stage B, 6.43 (3.37 to 12.24) for stage C, and 9.03 (5.12 to 15.95) for stage D. After inclusion of age and stage, the effect of residence in a `teaching' District was negligible with a hazard ratio of 0.96 (0.67 ± 1.38), P ˆ 0.81. Discussion Bias in the sample could arise in three ways. Incomplete data We were able to examine notes of only 263 of the 296. Of these, six had incorrect anniversary date and ®ve had the wrong cancer registered. Thus, 252 notes were available and eligible for study giving an 85% success rate. Some items were not completed on our record form by the tumour registrars: although this proportion missing from record sheet was low. (Table 2). Non-current data The data were not contemporary being sampled from 1994 and it is possible that treatment practices have since

Figure 3 Validity of Registry staging: Survival by Registry-imputed stage.

Prostate survival P Silcocks et al

160

changed. Retrospective data collection was necessary to ensure complete ascertainment and to allow some followup for validation. Notes which did not refer to patient's principal admission It is possible that the proportion of clinician-staged cases, and cases for which pathology reports were available, would have been higher if the notes referring to the patient's principal admission were used rather than the hospital from which the registration was made, as these are not always the same. On the other hand our study does give an indication of current feasibility and extent to which stage could be collected. TNM staging as such was impractical for prostate cancer, but we were able to adopt another recognised system, rather than having to devise a new scheme of our own. It is likely however that the differences in survival between stages we observed are less than the true differences Ð partly because if in doubt we `rounded down' cases to lower stages, and partly because of the well-known problem of clinical understaging; previously8 the only remedy for which is more thorough clinical investigation. It is also worth comparing our results with those of a large American survey9 in which 21.5% of white cases were stage A, 31.6% stage B, 17.7% stage C and 20% stage D. Our series has fewer stage C and more stage A cases. Sample size This was entirely adequate for demonstrating the effect of stage, as may be seen by inspection of the con®dence limits: even for the comparison of stage B with stage A the post hoc power to detect a hazard ratio of three was about 89% at a two-tailed P ˆ 0.05. Lessons learnt Personnel and ®nancial implications Data abstractors need adequate training, whether based in Trusts or at the Registry and there needs to be consideration as to who should collect this information. While Registry staff can do this, it doesn't follow automatically that this should be their responsibility. Registry-based clerical staff without medical or nursing training were able to collect prognostically valid information from 96% of the case-notes, despite their own misgivings and initial lack of con®dence. This was despite the proportion of cases in which explicit TNM and M categories were recorded in case notes being low. If Registry staff did this rather than clinical coders there would be a gain in consistency. Resource implications The average of 20 min per set of case notes might be slightly less in practice (because our study also included an audit check on the accuracy of basic standard data held at the Registry). Nevertheless even at 15 min per set of notes the cost for routine staging of cancers will be considerable; if clinical coders were to abstract this information for breast, cervix, colorectal cancers and melanomas (as is required in the core contract for Cancer Registries) the

annual salary cost alone for the Trent Region would be around £15 000. One way to reduce costs would be to ensure that a suitably responsible member of each clinical team recorded stage and its basis in standardised form in an easily accessible part of the notes, or in a compatible clinical database. This would provide greater and more accurate prognostic discrimination. Collection of stage information The point at which data are abstracted from case-notes needs reconsideration: case-notes are not always complete at discharge, coding and treatment over the six months after diagnosis will not be recorded unless surgical and treatment given as an in-patient or day case. Flexibility was needed in application of staging rules as in practice clinicians tend not to distinguish clinical from pathological TN and M categories. It was necessary to take the best-quality information available for each axis to produce an overall pro®le. Moreover some investigations are done before the anniversary date (based on the histological con®rmation) Ð for example tumour size may be assessed by rectal examination several days before the histology report is issued. We also found that bone scans were often performed several months after the original diagnosis Ð even with a slow growing rumour these cannot all be equivalent to a scan done within a fortnight of diagnosis. Hence, after consultation we adopted a window for stage-related investigations of two weeks before to two months after the anniversary date. This window is likely to differ between cancer sites and needs to be reviewed in order to produce a wider concensus. It is an issue which has not been addressed in the TNM manual. Acknowledgements Mr J Anderson FRCS (Consultant Urologist, Royal Hallamshire Hospital) for advice and comments; Mrs. Brenda Chattle, Mrs Joy Tingle, Ms Sarah Hethershaw & Mr Richard Powell (Tumour Registrars) for data collection. References 1 Calman KC, Hine D. A Policy Framework for Commissioning Cancer Services. (The Calman-Hine Report) Expert Advisory Group on Cancer. HMSO: London, 1995. 2 UICC International Union against Cancer. TNM Classi®cation of Malignant Tumours 5th ed. Springer-Verlag: Berlin, 1992. 3 Whitmore WF. Natural history and staging of prostate cancer. Urol Clin North America 1984; 2: 205 ± 220. 4 Cox DR. Regression models and life-tables (with discussion). J Roy Statist Soc 1972; B34: 187 ± 220. 5 Campos-Filho N, Franco EL. Microcomputer-assisted univariate survival data analysis using Kaplan-Meier life table estimators. Comp Meth Prog Biomed 1988; 27: 223 ± 228. 6 Campos-Filho N, Franco EL. A microcomputer program for multiple regression by Cox proportional hazards model. Comp Meth Prog Biomed 1990; 31: 81 ± 87. 7 User's Guide Microsoft EXCEL version 5 for Windows, Microsoft Corporation. 8 Epstein JI, Walsh PC, Carmichael M, Brendler CB. Pathologic and clinical ®ndings to predict tumour extent of nonpalpable (stage T1c) prostatic cancer. JAMA 1994; 271: 368 ± 374. 9 Murphy GP et al. The national survey of prostate cancer in the United States by the American College of Surgeons. J Urol 1982; 127: 928 ± 934.