Family Practice Advance Access originally published online on June 12, 2008
Family Practice 2008 25(4):215-220; doi:10.1093/fampra/cmn023
Identifying patients with a cancer diagnosis using general practice medical records and Cancer Registry data
a Centre for Primary Health Care and Equity, University of New South Wales, Sydney 2052, Australia
b Department of Primary Care and Public Health, North Wales Clinical School, Cardiff University, Wrexham Technology Park, Wrexham LL13 7YP
c Academic Unit of Primary Care, Leeds Institute of Health Sciences, Charles Thackrah Building, University of Leeds, 101 Clarendon Road, Leeds LS2 9JL
d Hull York Medical School, University of York, Heslington YO105DD, UK
e RAND Corporation, 1776 Main Street, Santa Monica, CA, USA
f Cancer Research UK Clinical Centre, St James's University Hospital, Beckett Street, Leeds LS9 7TF, UK
Correspondence to Shane W Pascoe, Centre for Primary Health Care and Equity, University of New South Wales, Sydney 2052, Australia; E-mail: shanewpascoe{at}hotmail.com
Received 1 August 2007; Revised 4 February 2008; Accepted 21 April 2008.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Background. The medical records of patients with cancer need to accurately record diagnoses for professionals to provide quality care.
Aims. (i) To develop a methodology which identifies medical records of patients with a cancer diagnosis. (ii) To describe the effectiveness of search strategies to identify all patients in primary care with a cancer diagnosis compared with a diagnosis identified by a Cancer Registry.
Methods. The design of the study was a retrospective analysis of primary care medical records. Five general practices were recruited in the UK. The completeness and correctness of searches were measured and compared both within the practices and compared with a diagnosis identified by a Cancer Registry.
Results. One in five of all primary care patients with cancer was not identified when a search for all patients with cancer was conducted using electronic codes for malignancy. One in five patient records with an electronic code for a malignancy that was confirmed by registration with the Cancer Registry actually lacked the necessary documentation to verify the cancer type, date of diagnosis or any other aspect of the malignant condition. Overall, electronic codes for cancer in these medical records have a poor level of completeness (29.4%) and correctness (65.6%) when compared with the Cancer Registry.
Conclusions. The electronic codes in five general practices were not able to identify all patients on the practice lists with a cancer diagnosis. Practices will only be able to comply with guidelines and meet quality targets if they can identify all of their current patients with a cancer diagnosis and will require information from a Cancer Registry in order to do this.
Keywords. Primary care, cancer, medical records.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
The need for a comprehensive UK National Strategy to manage patients with cancer led to the development of the National Health Service (NHS) Cancer Plan in 2000.1 Included was an initiative to develop a clinical data set of all patients with cancer in primary care.1 The new general practice contract (2006) includes quality indicators for cancer (with associated financial incentives) as part of the quality and outcomes framework (QOF). All patients with cancer must be identified and require clinical and psychosocial review.2 An up to date clinical data set is essential to meet this quality indicator.
Many chronic illnesses are now managed within primary care, requiring comprehensive call and recall systems. Despite this, no studies have quantified the completeness or correctness of recording cancer diagnoses in primary care. Under the new contract, no reporting and verification indicators have been described; hence, practices cannot establish whether their registers are complete. By addressing limitations to the use of electronic codes in primary care to assess quality, one small study found that QOF higher scores were not related to adherence to Royal College of Physician guidelines for stroke management.3
Given the problems with paper record systems and the lack of accurate transfer of information,4,5 one hypothesis developed by the authors questions whether current clinical management systems in general practices are able to identify all patients with cancer on their lists. The aims of the study were to develop a methodology that identified patients with an established diagnosis of cancer in general practice and secondly to describe the effectiveness of a search strategy used to identify all patients with a registered cancer diagnosis by comparing with an established data set (namely a regional Cancer Registry).
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Setting
Six general practices were recruited through an advertisement distributed by Primary Care Trust (PCT) Primary Care Cancer Lead in each of the five Leeds PCTs. Five practices were then purposely selected to represent different practice list sizes and levels of computerization. By chance, all the study practices used the Egton Medical Information Services system, in common with 55% of UK GPs.6 Like other computer databases used by the NHS, EMIS manages clinical information using the Read classification system.
The study was carried out in four stages within each practice:
- Various searches of the electronic records generated a list of patients aged >18 years on December 31, 1989 with a ‘suspected’ cancer diagnosis from 1990 to 1999 inclusive. The time range was chosen to provide a sample powerful enough to draw generalizable conclusions.
- A manual search of paper medical records was used to verify or refute the cancer diagnosis.
- Search strategies were analysed in order to determine the yield from each search.
- Combined searches were compared with data from the regional Cancer Registry [Northern and Yorkshire Cancer Registry and Information Service (NYCRIS)].
Three data extractors were used. Each had a high level of experience working in primary care and were trained in data abstraction. Sample records were used in training the data extractors. The sample size calculation was based on 250 patients with cancer in Practice 1 and 250 matched controls allowing us to detect odds ratios of at least 2.2, based on 94% power at 5% significance between the cancer and control patients. To allow for the fact that there are multiple primary outcome measures (adjusted for five), the power based on 250 patients in each group is 83% at a 1% significance level.
Stage 1. Search strategy identifying medical records of patients with a suspected cancer diagnosis
The initial identification of patients with a suspected cancer diagnosis was carried out using multiple searches in the five practices during 2002 and 2003. The methods reflect pragmatic searches completed by practices to identify patient subgroups during everyday practice. Records of patients with a suspected diagnosis of cancer were identified by:
- (i) Searches of diagnostic codes for cancer in electronic records with benign codes excluded (for all patients listed on the practice computer system; whether alive, deceased or removed from the list).
- (ii) Searches of diagnostic codes for cancer-related activity (e.g. oncology referrals, tamoxifen prescriptions, mastectomy, positive cervical cytology reports).
- (iii) GPs recall patients diagnosed with cancer and list them (all practices were requested to provide these data; however, only two practices provided the lists).
- (ii) Searches of diagnostic codes for cancer-related activity (e.g. oncology referrals, tamoxifen prescriptions, mastectomy, positive cervical cytology reports).
Stage 2. Verification of a cancer diagnosis
The electronic and paper records of suspected cancer patients were then examined to confirm or refute a new diagnosis of cancer in the period 1990–1999 inclusive. The end date was retrospective at the inception of the study as the authors also wished to study the records of this cohort of patients for a period of time after diagnosis. For example, for a patient with a breast cancer diagnosis in 1999, data were abstracted for a subsequent 2-year period. These results will be reported in a subsequent article. A record was defined as having a cancer diagnosis if it contained either a letter from a consultant reporting a cancer diagnosis or a pathology report diagnosing cancer. This was the measure against which all suspected diagnoses were verified. Records of deceased patients were requested from the West Yorkshire Central Services Agency (WYCSA). Records of patients who had moved practices were unavailable to the investigators.
Stage 3. Confirmation from each search and in combination compared with the Cancer Registry
A list was created which identified all patients registered on the practice list at the beginning of the data collection period (December 31, 1989) and then adding all new patients who joined the list over the next 10 years. This new list of all patients present during any part of the selected time period was searched and data extracted (e.g. names, birth dates and NHS numbers).
The list held data for all patients suspected of having a cancer diagnosis. This list could be compared with Cancer Registry data (NYCRIS). NYCRIS has collected data from multiple sources since 1975.7,8 Histological reports are often, but not exclusively, the source for notification of a cancer diagnosis.9 Other sources included radiotherapy records, screening services, independent hospitals, specialist tumour registries, hospices and death certificates.9 The five practice lists were compared with data from NYCRIS to produce lists of patients at each practice identified by both sources as having a cancer diagnosis. This process also identified patients whose cancer diagnoses were registered with NYCRIS but not with the practice and vice versa. NYCRIS cohort data were extracted using postcodes for the five Leeds PCTs. As one of the practices included patients from PCTs other than Leeds, the boundaries were extended to include Airedale, Craven, Harrogate and Rural District PCT. Both the practice lists and the NYCRIS data were loaded onto a patient matching database. The algorithm identifies the NHS number as the first data item and compares the practice list and NYCRIS list for matches. The next data items are the date of birth followed by the surname, which were used for non-matching or obsolete NHS numbers. The confirmed diagnostic yields from individual and combined search strategies were calculated. The percentage of patients correctly identified was then calculated.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
The organization of records in the study practices was complex, as each practice adapted to the differing demands of new information management systems during the 1990s. Data recording in these practices can be characterized as dynamic and with no systematic measurement of data quality. The format of paper records differed between the practices; one practice used A4-sized paper records while the remainder used traditional A5 envelopes (so-called ‘Lloyd George’ notes), to hold the consultation history, letters and test results. Incoming correspondence was folded to fit into the envelope. In these practices, when a record held a large number of entries, separate files were used to archive any information regarding past medical history. More recent paper records were kept in a separate medical records filing room. Two of the five practices had not completed the transfer of paper summaries to electronic records. In transferring information from the paper record to the electronic record and managing the day to day data recording, some training was provided to staff by the practices. The quality of the training provided and the range of skills taught were not measured by the practices. Practice managers described the training provided by practices as ad hoc, and was often dependant on the motivation of an individual GP.
Search strategy to generate lists of patients with a suspected cancer diagnosis.
A total of 1741 records of patients with a suspected cancer diagnosis were identified from the patient lists. The numbers for each practice are shown in Table 1.
|
Verification of a cancer diagnosis
In all, 1137 (65%) of the paper records of suspected cancer patients were available for analysis, varying between practices from 61% to 80%. In all, 297 records could not be assessed because they were unavailable from WYCSA after the death of the patient (Fig. 1); this number is inexplicably high because records are not usually destroyed until 8–10 years after the patient's death. In all, 257 records were not reviewed because patients had moved practice. In all, 48 records of patients with a suspected diagnosis could not be found after searching the study practices, although they were still registered. Two records were excluded, as the patients were under 18.
|
The yield from each search strategy and combination of searches
After reviewing the records, 742 of the 1137 (65%) patients were confirmed as having a diagnosis of cancer, the percentage of confirmed from suspected cases in each practice ranging from 61% in Practice 1 to 71% in Practice 3 (Fig. 1).
Searches of electronic codes for a malignancy. No single search of electronic records identified all patients with a possible diagnosis of cancer confirmed in the paper record. A combination of Read codes for malignancy, codes for abnormal smear results and prescriptions for tamoxifen identified 96.3% of records containing a cancer diagnosis (Table 1). One in five patients (19.3%) whose paper records confirmed a malignancy had no electronic code for malignancy in the electronic record; this reached 57% in one practice. Each search strategy except radiotherapy identified at least one possible cancer patient not identified by any other search.
Non-cancer diagnoses. Of the 395 patients who were on the suspected cancer list but whose cancer diagnosis could not be confirmed, 194 (49%) were found to have a benign condition, 89 (23%) had a confirmed cancer diagnosis but it had occurred outside the time period of the study and 102 (26%) could not be confirmed because the necessary documentation was not in their record. A summary of records identified in each practice containing a confirmed cancer diagnosis can be found in Figure 1.
Comparison with Cancer Registry data. The combined search strategies failed to identify all patients registered with the Cancer Registry. The sensitivity rates (completeness) and positive predictive values (correctness) for the combined search strategy were examined and did not vary as a function of the year of diagnosis. Overall, electronic codes for cancer in these medical records have a poor level of completeness (29.4%) and correctness (65.6%) when compared with the Cancer Registry (Table 1). Routinely collected data stored in the paper record were not able to identify a significant minority of patients with a verified diagnosis of cancer (60/647—9.3%; Table 2). Of the 60 records verified by the Cancer Registry, but without a confirmed diagnosis in the medical record, 13 had a Read code for malignancy (21.7%; Table 2). Almost three quarters of records of patients with breast cancer in the Registry that did not contain an electronic morbidity code for malignancy could be identified through a prescription for tamoxifen.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Summary of main findings
Electronic searches of codes for malignancy identify the largest number of records containing a cancer diagnosis. However, these codes failed to identify one in five patients with a cancer diagnosis. Patients not identified by a code for malignancy were identified through other searches, such as a prescription for tamoxifen. Almost half of the records with an electronic code for malignancy and without confirmation in the paper record had a benign neoplasm confirmed in the paper record. Benign conditions may have been incorrectly coded as malignancies when entering data from incoming correspondence. A similarly high rate of transcription errors has been reported elsewhere.4 Only a small percentage of records with an electronic code for malignancy but without confirmation in the paper record were identified by the Cancer Registry (7.7%; 13/171).
The most effective method of establishing a list of patients with a cancer diagnosis may be the use of the multi-modal search method adopted in the study combined with a systematic cross-checking of the paper record belonging to each patient identified. External validation with, for example, the regional Cancer Registry will improve the completeness and accuracy of a practice-based registry. Of utmost importance, and now probably widely in operation, is to correctly identify and code all new diagnosis.
Strengths and limitations of the study
The study was undertaken in five general practices of varying sizes, drawing on different populations in the Leeds area, and the abstraction of data was as comprehensive as possible. One-third of the paper records of patients with a suspected cancer diagnosis could not be assessed, predominantly due to deaths and movement out of the practice area. However, this does not affect the validity of the findings for patients still registered with the practices, for whom the findings have most importance. The generalizability of the findings, given the practices involved, is an important issue. To an extent, all practices are atypical. It could be argued that these practices were more atypical than others given their high level of involvement in PCT activities, and information technology use. Their coding may therefore be better than other practices. Lastly, the age of the data is an issue. We accept that the quality of coding may now be better than at the time of this retrospective study, particularly with respect to correct and complete recording of new diagnoses contemporaneously. However, a clear record of a cancer diagnosis should be easily accessible in all cancer survivors’ records, as it is of huge clinical importance. Cancer survivors have complex needs relating to their cancer, much of which is delivered in primary care.10–12 This includes monitoring for recurrence and for second cancers, monitoring for the side effects of treatment and the detection and treatment of psychosocial issues.
Comparison with existing literature
Good quality electronic records can provide support for the physician, for the benefit of patient care.13–18 The findings of this study suggest that a significant number of patients are not known from their records to have had a previous diagnosis of cancer. Recent studies have found that achievements of QOF indicators are poorly related to the number of patients registered,19 health gain20 or to patient satisfaction.21 Other research has found that the majority of practices are achieving all their clinical indicators for cancer (91.7% of practices) and are reimbursed accordingly.22 A search strategy similar to the one used in this and previous studies23 highlights the importance of using multiple sources such as prescriptions for tamoxifen, radiotherapy reports and a mastectomy/lump/biopsy code to identity incident cases of cancer.
Implications for clinical practice
These findings show that a clinical system used extensively in general practice is not able to identify all patients with a cancer diagnosis. Over the 10-year period of the study, the completeness and correctness of the search strategy used did not improve data quality, suggesting the transition to electronic records requires audit in general practices. Even within the practices that have summarized data to electronic records, poor data quality still remained an issue. To identify the many patients with cancer in the five general practices involved in this study, a large number of paper records had to be manually searched. The present system whereby administrative members of the primary health care team enter most data into medical records can produce errors (e.g. benign conditions coded as malignancies), which are not identified if there are no auditing procedures in the practice. In addition to the need for audit, patients are presenting opportunistically each day, and it is likely that poor quality care will result from no mention of a recent cancer diagnosis on their medical record during these consultations. A more systematic process to identify patients is therefore likely to lead to better care. This process could include the use of Cancer Registry data described by this study.
|
Declaration |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
Funding: Pan-Leeds Data Quality Group NE Leeds PCT.
Ethical approval: The St James Hospital Leeds Local Research Ethics Committee provided ethics approval for the study.
Conflicts of interest: None.
|
Acknowledgments |
|---|
We would like to thank the practices involved in the study and Joyce Evans and Pauline Nelson for assisting with data collection. The support offered by NYCRIS was substantial and without which the project could not have proceeded. Thanks to Nat Wright and Brenda Leese for comments on a draft of the paper.
|
Notes |
|---|
Pascoe SW, Neal RD, Heywood PL, Allgar VL, Miles JNV and Stefoski-Mikeljevic J. Identifying patients with a cancer diagnosis using general practice medical records and Cancer Registry data. Family Practice 2008; 25: 215–220.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionDeclarationReferences |
|---|
1 Department of Health. The NHS Cancer Plan (2000) London: Department of Health.
2 Department of Health. New GMS Contract 2006/07 (2006) London: Stationary Office.
3 Williams PH, de Lusignan S. Does a higher ‘quality points’ score mean better care in stroke? An audit of general practice medical records. Infom Prim Care (2006) 14:29–40.
4 Hippisley-Cox J, Pringle M, Cater R, et al. The electronic patient record in primary care—regression or progression? A cross sectional study. Br Med J (2003) 326:1439–1443.
5 Sullivan F, Wilson P. The completeness and accuracy of patient record transfer between practices. Health Bull (1997) 55:16–19.
6 Egton Medical Information Services. http://www.emis-online.com. (last accessed 2004).
7 Department of Health. Cancer Registration (2003) http://www.doh.gov.uk (last accessed 1 January 2003).
8 Dickinson HO, Salotti JA, Birch PJ, Reid MM, Malcolm A, Parker L. How complete and accurate are cancer registrations notified by the National Health Service Central Register for England and Wales? J Epidemiol Community Health (2001) 55:414–422.
9 Northern and Yorkshire Cancer Registry and Information Service. Registration of Skin Cancer in Yorkshire. A Study into the Completeness and Validity of Cancer Registry Data (2001) Leeds: NYCRIS.
10 Pascoe SW, Neal RD, Allgar VL, Selby P, Wright P. Psychosocial care for cancer patients in primary care—recognition of opportunities for cancer care. Fam Pract (2004) 21:437–442.
11 Campbell NC, MacLeod U, Weller D. Primary care oncology: essential if high quality cancer care is to be achieved for all. Fam Pract (2002) 19:577–578.
12 Allgar VL, Neal RD. General practitioners’ management of cancer: secondary analysis of data from the National Survey of NHS Patients: Cancer. Eur J Cancer Care (2005) 14:409–416.[CrossRef]
13 Whitelaw FG, Nevin SL, Milne RM, Taylor RJ, Taylor MW, Watt AH. Completeness and accuracy of morbidity and repeat prescribing records held on general practice computers in Scotland. Br J Gen Pract (1996) 46:181–186.[Web of Science][Medline]
14 Purves IN. PRODIGY: implementing clinical guidance using computers. Br J Gen Pract (1998) 48:1552–1553.[Web of Science][Medline]
15 Morris AD, Boyle DI, MacAlpine R, et al. The disease audit and research in Tayside Scotland (DARTS) study: electronic record linkage to create a diabetes register. DARTS/MEMO collaboration. Br Med J (1997) 315:524–528.
16 LaPorte RE, Akazawa S, Hellmonds P, et al. Global public health and the information superhighway. Br Med J (1994) 308:1651–1652.
17 Hobbs FDR, Hawker A. Computerized data collection: practicability and quality in selected general practices. Fam Pract (1995) 12:221–226.
18 Dovey SM, Tilyard MW. The computer research network of the Royal New Zealand College of General Practitioners: an approach to general practice research in New Zealand. Br J Gen Pract (1996) 46:749–752.[Web of Science][Medline]
19 Guthrie B, McLean G, Sutton M. Workload and reward in the Quality and Outcomes Framework of the 2004 general practice contract. Br J Gen Pract (2006) 56:836–841.[Web of Science][Medline]
20 Fleetcroft R, Cookson R. Do the incentive payments in the new NHS contract for primary care reflect likely population health gains? J Health Serv Res Policy (2006) 11:27–31.
21 Brown C, Lilford R. Cross sectional study of performance indicators for English Primary Care Trusts: testing construct validity and identifying explanatory variables. BMC Health Serv Res (2006) 6:81.[CrossRef][Medline]
22 Doran T, Fullwood C, Gravelle H, et al. Pay-for-performance programs in family practices in the United Kingdom. N Engl J Med (2006) 355:375–384.
23 Kaye JA, Derby LE, del Mar Melero-Montes M, Quinn M, Jick H. The incidence of breast cancer in the General Practice Research Database compared with national cancer registration data. Br J Cancer (2000) 83:1556–1558.[CrossRef][Web of Science][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  A. Majeed, J. Car, and A. Sheikh Accuracy and completeness of electronic patient records in primary care Fam. Pract., August 1, 2008; 25(4): 213 - 214. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||