Canadian Forces Evaluation of the EPINATO Health Surveillance System in Bosnia-Herzegovina

Transcription

1 MILITARY MEDICINE, 171, 10:955, 2006 Canadian Forces Evaluation of the EPINATO Health Surveillance System in Bosnia-Herzegovina Guarantor: Maureen T. Carew, MD FRCPC Contributors: Jean L. Wilson, MHSc* ; Maureen T. Carew, MD FRCPC*; Barbara A. Strauss, RN MSc* The Canadian Forces (CF) adopted the EPINATO surveillance system in 1996 to monitor disease and injury morbidity in deployed settings. The Directorate of Force Health Protection, CF Health Services Group initiated an evaluation of EPINATO in Task Force Bosnia-Herzegovina in August Two methods were used to assess coding reliability: a chart audit and Sick Parade Register review. Stakeholder interviews were conducted evaluating data flow, reporting structure, and key system attributes. Reliability (, 95% confidence interval) was good in 4 of 24 categories sexually transmitted diseases, 0.75 (0.50, 1.00); eye disorders, 0.51 (0.15, 0.88); ears/nose/ throat, 0.51 (0.33, 0.69); lower respiratory infections, 0.49 (95% confidence interval 0.34, 0.65) but otherwise was poor. EPINATO is not an effective, reliable tool for CF deployment health surveillance. An improved health surveillance system is required to ensure disease and injury aberrations are detected and optimal preventive programs and policies are in place for deployed CF military members. Introduction ealth surveillance systems are an essential element in disease prevention and control for military forces that face a H continuum of operational scenarios including war, peacekeeping, potential terrorist use of biological weapons, and natural disasters. The ongoing, systematic collection, analysis, interpretation, and dissemination of health events provides critical health status information about serving military personnel which can assist in early identification of new or emerging public health threats. In 1996, the Surgeon Generals of the North Atlantic Treaty Organization (NATO) nations approved the EPINATO surveillance system for use among NATO member countries with this vision in mind. The Canadian Forces (CF) adopted EPINATO for deployment health surveillance in The purpose of EPINATO as described in the NATO Standardization Agreement (NATO STANAG 2235, Edition 1) was to monitor disease and injury problems in military units before they limited mission effectiveness and guide the development of prevention and control programs and policies. At the NATO level, the system would provide standardized multinational health surveillance data for comparison between countries participating in the same theater. *Canadian Forces Health Services Group Headquarters Directorate of Force Health Protection, Ottawa, Ontario, K1A 0K6, Canada. Public Health Agency of Canada Canadian Field Epidemiology Program, Ottawa, Ontario, K1A 0L2, Canada. Presented at NATO Medical Surveillance and Response: Research and Technology Opportunities and Options, April 19 21, 2004, Budapest, Hungary, and at the Committee of the Chiefs of Military Medical Services in NATO (Preventive Medicine Working Group) Meeting, April 22, 2004, Budapest, Hungary. This manuscript was received for review August The revised manuscript was accepted for publication in March Reprint & Copyright by Association of Military Surgeons of U.S., Before the implementation of EPINATO, two pilot studies were conducted in Bosnia- Herzegovina to evaluate the J95 surveillance system (containing 32 codes), EPINATO s British predecessor. Jefferson et al. 1 conducted a coding validity and reliability study in three British military units. A 65% concordance was reported between field physicians, the gold standard (expert), and the recoding of events among three physicians. Percent agreement between each of three sites to the gold standard was high at an average of 85%. Owen et al. 2 piloted the system in several secondary care facilities and used the system to analyze health events, work-related burden, and health care utilization. Both studies recommended simplifying the classification structure and enhancing personnel training to improve data quality and system acceptability. Both studies also concluded that the J95 surveillance system was a feasible way to systematically collect data, carry out population surveillance, and aid in future resource planning. 1,2 In August 2003, the Directorate of Force Health Protection (D FHP), CF Health Services Group (CF H Svcs Gp HQ) evaluated the EPINATO surveillance system as used in Task Force Bosnia- Herzegovina (TFBH). The purpose of the evaluation was to determine the effectiveness of EPINATO as a CF deployment health surveillance system. The specific evaluation objectives were to: 1. Identify the original objectives of the EPINATO surveillance system. 2. Describe the flow of data within the CF and NATO reporting structures. 3. Document the analysis, interpretation, and dissemination (operational process) of data from EPINATO by the CF and NATO. 4. Determine whether EPINATO information guided prevention programs and health policies within the CF. 5. Assess the following system attributes: data quality (reliability/validity), timeliness, acceptability, and representativeness. Methodology The evaluation was carried out between July 2003 and January 2004 (Roto 12) and was structured according to the Centers for Disease Control and Prevention Guidelines for Evaluating Public Health Surveillance Systems. 3 A coding reliability study and key stakeholder interviews were conducted to describe the system and its key attributes (data quality (reliability/validity), timeliness, acceptability, and representativeness). Reliability was defined as the consistency of coding the same diagnosis between different coders. 4 Validity was defined as the system s ability to measure outcomes that it had originally intended to measure. 5 Representativeness was defined as the system s ability to accurately describe the 955

2 956 CF EPINATO Evaluation in Bosnia-Herzegovina occurrence of health events over time distributed by person and place within the population. 3 TABLE I EPINATO CODES AND DESCRIPTIONS Code No. Description 1 Intestinal infectious disease 2 Syphilis and other STDs 3 Other infectious and parasitic diseases 4 Alcohol and drug abuse and dependencies 5 Mental disorders 5.1 Stress reactions 6 Eye disorders 7 Disorders of the ear, nose, and throat 8 Diseases of the lower respiratory tract 9 Disease of teeth and oral cavity 11 Disease of the digestive system 12 Gynecological diseases (including pregnancy) 13 Dermatological problems 14 Internal derangement of the knee 15 Dorsopathies 16 Other musculoskeletal diseases 17 Medical complications 18 Other diseases 19 Injury due to road traffic accidents 20 Injury due to (military) training 21 Injury due to sports 22 Injury due to war/operations 23 Other injury 24 Climatic injury (heat and cold) 25 NBC indicators Coding Reliability Study In August 2003, two FHP epidemiologists completed a retrospective chart review to assess data reliability at two CF unit medical stations (UMS) in TFBH (Camp A and Camp B). These camps were chosen because they had the largest populations and because their proximity to one another facilitated on-site visits. A convenience sample was used in theater in place of the planned random sample within each EPINATO category due to a logistical problem of accessing health records in the UMS. To simplify the sampling process, the health records of the patients with the most recent visits were selected. All visits within the patients health records were then coded according to the 25 EPINATO categories (Table I). The charts included a total of 128 visits (103 from Camp A and 25 from Camp B) which were divided among the two FHP reviewers who assigned EPINATO codes as defined in the Stabilization Force-CJ4 Medical Epidemiology Rules of the Game document (Sarajevo, March 1998). FHP reviewer no. 1 assigned codes for 99 of 128 (77%) of UMS visits and the remaining 29 of 128 (23%) were coded by reviewer no. 2. The FHP team was blinded to the EPINATO codes previously assigned by the UMS staff. Similarly, EPINATO codes were assigned by UMS staff before the evaluation negating the potential for a Hawthorne effect. FHP and UMS EPINATO codes were compared and reliability for individual categories was assessed using Cicchetti-Allison weighted and 95% confidence intervals (CIs) using SAS software, version 9.1 (SAS Institute, Inc., Cary, North Carolina). values of 1.00 and 1.00 reflected perfect disagreement and perfect agreement beyond chance, respectively ([ evaluation scale: 0.75 excellent, good, marginal reproducibility]. 6 A priori weights were assigned for each EPINATO code and the weighting matrix is described in Table II. An aggregated, unweighted was calculated to determine overall reliability (EPINATO codes 1 25). An electronic version of the complete Sick Parade Register (SPR) from the largest camp was provided to FHP at the end of Roto 12 which contained information on the chief complaint, diagnosis, treatment, and UMS EPINATO code for all 1,600 visits to the UMS between March 29, 2003 and September 30, A sensitivity analysis was done using all disease-based visits to determine the influence of the nonrandom sampling of health records in theater. Injury visits were excluded from the SPR analysis because the mechanism of the injury (e.g., sports, training, road traffic accident) was not identified on the SPR. A physician epidemiologist from FHP assigned EPINATO codes according to the Rules of the Game document and was blinded to codes previously assigned by the UMS staff. UMS and FHP codes were compared using a Cicchetti-Allison weighted and 95% CIs using the same a priori weights described in Table II (codes per rater). An overall was not calculated since injury codes were excluded from the SPR analysis. The SPR was also examined to determine the proportion and characteristics of visits captured in nonspecific categories (i.e., no. 3 Other Infectious & Parasitic Diseases, no. 16 Other Musculoskeletal, no. 18 Other Diseases, no. 23 Other Injuries). Stakeholder Interviews A standardized questionnaire was developed with open-ended questions to determine EPINATO s original objectives, data flow, TABLE II RANK ORDERED CATEGORIES FOR WEIGHTED ANALYSIS Code First Ranked Categories Second Ranked Categories Third Ranked Categories , 2 7, 8, , , , , 2, 3, 4, 5, 5.1, 6, 7, 8, 9, 11, 12, 13, 14, 15, , , ,

3 CF EPINATO Evaluation in Bosnia-Herzegovina reporting structure, operational process, data utilization, content validity, timeliness, acceptance, and representativeness. Stakeholder interviews were conducted (with UMS and NATO personnel) between August 2003 and December 2003 in three areas of operation: theater (TFBH), garrison (CF H Svcs Gp HQ), and NATO (Supreme Headquarters Allied Powers Europe (SHAPE)). The questionnaire was administered face to face in TFBH and by or telephone for CF H Svcs Gp and SHAPE participants. The site visit to TFBH provided an opportunity to observe the implementation of EPINATO in two CF camps (Camps A and B) and one multinational camp (Camp C). Results Coding Reliability Chart Audit In the chart audit, UMS EPINATO codes were missing for 16% (21 of 128) of visits, leaving 107 observations for analysis. The unweighted analysis found an overall 40% agreement between FHP and UMS coding ( 0.40 (95% CI, )). Weighted and 95% CIs are presented in Table III for each of the 18 EPINATO categories for which events were observed. Categories with more specific clinical presentations (no. 2, sexually transmitted diseases (STDs); no. 6, eye disorders; no. 7, ears/nose/ throat (ENT); no. 8, lower respiratory) had higher values of 0.75 (95% CI 0.50, 1.00), 0.51 (95% CI, 0.15, 0.88), 0.51 (95% CI, 0.15, 0.88), and 0.49 (95% CI, 0.34, 0.65), respectively. However, for the remaining EPINATO codes, reliability estimates 957 were very low with values ranging from 0.16 for mental disorders to 0.29 for gynecological visits. Coding Reliability SPR A total of 1,600 visits for EPINATO disease-related categories (nos. 1 18) was included in the SPR analysis. Results of the weighted analysis are presented in Table IV. There were no visits for code 25 (nuclear, biological, chemical indicators) and the sample size was sufficient to calculate values for each remaining disease category. Reliability estimates in the SPR were generally lower when compared to the same category in the chart audit. Only two categories (no. 7-ENT and no. 8-lower respiratory) were found to have moderate reliability with of 0.55 (95% CI 0.51, 0.60) for ENT and 0.47 (95% CI 0.42, 0.51) for lower respiratory codes. values for the remaining 13 disease-related EPINATO codes indicated very poor reliability with ranging from 0.15 for dermatological problems (no. 13) to 0.22 for stress reaction (no. 5.1). Twenty-five percent (401 of 1,604) of visits in the SPR were coded under other nonspecific categories. Review of these categories found several diagnoses (e.g., no. 1 Intestinal infectious disease, no. 7 Disorders of the ENT, and no. 13 Dermatologic problems) that could have been captured under more specific EPINATO codes. As well, important clinical entities such as cardiovascular, urological, and neurological visits could not be uniquely identified using the EPINATO tool. According to the Rules of the Game document, these conditions are grouped in the other diseases category. TABLE III CHART AUDIT CAMPS A AND B CODING RELIABILITY FOR EPINATO CATEGORIES Code EPINATO Category Total No. UMS No. FHP Weighted (95% CI) 1 Intestinal infectious diseases ( 0.04, 0.44) a 2 Syphilis and other STDs (0.50, 1.00) a 3 Other infectious and parasitic diseases b b b b 4 Alcohol and drug abuse ( 0.18, 0.18) a 5 Mental disorders ( 0.48, 0.19) 5.1 Stress reaction b b b 6 Eye disorders (0.15, 0.88) a 7 Disorders of the ear, nose, and throat (0.33, 0.69) 8 Diseases of the respiratory system (0.34, 0.65) 9 Diseases of the teeth and oral cavity b b b b 11 Diseases of the digestive system ( 0.39, 0.23) 12 Gynecological (including pregnancy) ( 0.03, 0.61) a 13 Dermatological problems ( 0.49, 0.18) 14 Internal derangement of the knee ( 0.15, 0.20) a 15 Dorsopathies ( 0.07, 0.41) a 16 Other musculoskeletal diseases ( 0.02, 0.38) a 17 Medical complications b b b b 18 Other diseases ( 0.17, 0.11) 19 Injuries due to road traffic accidents b b b b 20 Injuries due to training ( 0.22, 0.16) a 21 Injuries due to sports (0.02, 0.45) 22 Injuries due to war/operations b b b b 23 Other injuries-except when due to road ( 0.21, 0.05) a traffic accidents, training, sports, or war 24 Climatic injury (heat and cold) ( 0.11, 0.46) a 25 Nuclear, biological, chemical indicators (ops only) b b b b a Cell 5. b No events coded in this category.

4 958 CF EPINATO Evaluation in Bosnia-Herzegovina Stakeholder Interviews The following section presents the findings of the stakeholder interviews completed for a total of 20 personnel engaged at various levels of the EPINATO surveillance system (13 in theater, 4 in garrison, 3 NATO). EPINATO Objectives# No formal documentation was identified that defined the original objectives, planning, or implementation for the 1996 EPINATO surveillance system. Therefore, D FHP developed CF deployment surveillance objectives to guide the evaluation (Table V). Data Flow/Reporting Structure Stakeholder interviews documented the coding process as outlined in Table VI. The intended EPINATO reporting structure for TFBH is presented in Figure 1. Codes compiled at all CF camps were sent to the National Medical Liaison Officer at the CF Task Force Headquarters. Data were also collated and reported to CF Health Svcs Gp in Ottawa through the Deputy Chief of Staff Medical Operations. Simultaneously, the same report was sent to the Brigade Surgeon at the Multinational Brigade Headquarters Northwest in Banja Luka. Combined multinational reports were then forwarded to the Stabilization Force Headquarters (SFOR), Sarajevo, NATO Headquarters for the Balkans, Naples, and finally SHAPE, Belgium. Operational Process At the time of the evaluation, there was no analysis, interpretation, or dissemination of information collected at any level of the reporting structure (in theater, in garrison, and NATO). Stakeholders indicated that this process had ceased due to a lack of clear roles and responsibilities, no allocated resources, and poor data quality. Utilization of Data Because there was no analysis, interpretation or dissemination of information at any level, EPINATO data were not used for TABLE V CF DEVELOPED DEPLOYMENT SURVEILLANCE OBJECTIVES Surveillance Objectives 1 To monitor the incidence of communicable diseases among deployed CF members (based on clinical diagnoses) on a daily, weekly, annual, and end-of-roto basis, in order to detect disease outbreaks and to guide prevention and control programs 2 To monitor the incidence of injuries (battle and non-battle related) among deployed CF members on a weekly, endof-roto, and annual basis, in order to detect injury clusters and to guide prevention and control policies and programs 3 In addition to monitoring infectious respiratory diseases (under objective 1), the deployment surveillance system will monitor the incidence of non-infectious acute respiratory diseases (such as allergies and asthma) on a weekly, endof-roto, and annual basis, to detect disease clusters and guide prevention and control programs 4 To monitor on a weekly, end-of-roto, and annual basis, the incidence of health outcomes that may be related to exposure to environmental risk factors on deployment 5 To monitor on a weekly, end-of-roto, and annual basis, the incidence of mental illness among deployed CF members 6 To monitor on a weekly, end-of-roto, and annual basis, the severity of illnesses among deployed CF members by capturing the disposition of patients seen in the UMS (i.e., hospitalizations) 7 To provide on a monthly, end-of-roto, and annual basis a snapshot of the demand for medical specialist services on deployment health prevention and promotion programs, health policy development or resource allocation by the CF. Content Validity Health personnel described the classification rules and case definitions as confusing and vague. They indicated that a TABLE IV CAMP A ELECTRONIC SICK PARADE REGISTER (SPR) CODING RELIABILITY FOR EPINATO CATEGORIES Code EPINATO Disease Category Total n No. UMS No. FHP Weighted a (95% CI) 1 Intestinal infectious diseases ( 0.09, 0.07) a 2 Syphilis and other STDs ( 0.17, 0.32) 3 Other infectious and parasitic diseases ( 0.09, 0.00) a 4 Alcohol and drug abuse ( 0.05, 0.03) a 5 Mental disorders (0.15, 0.32) 5.1 Stress reaction (0.16, 0.29) a 6 Eye disorders (0.14, 0.39) 7 Disorders of the ear, nose, and throat (0.51, 0.60) 8 Diseases of the respiratory system (0.42, 0.51) 9 Diseases of the teeth and oral cavity (0.12, 0.23) 11 Diseases of the digestive system ( 0.10, 0.13) 12 Gynecological (including pregnancy) ( 0.33, 0.10) 13 Dermatological problems ( 0.21, 0.09) 17 Medical complications ( 0.50, 0.35) a 18 Other diseases 1, (0.06, 0.15 Total visits examined b (0.52, 0.59) a Cell 5. b Total number of visits for which there was an EPINATO disease category (injuries excluded).

5 CF EPINATO Evaluation in Bosnia-Herzegovina TABLE VI INTENDED CODING PROCESS: BOSNIA-HERZEGOVINA Steps Description 1 A CF member presents to the UMS with disorder/injury event 2 A diagnosis is made by a medical technician, physician assistant, and/or medical officer 3 A clerk assigns one of the 26 EPINATO codes per diagnosis. All codes are defined in the rules of the game document produced by the Stabilization Force (SFOR)-CJ4 Medical Epidemiology developed in March Visit status (e.g., initial or follow-up) and disposition (e.g., light duty days, days off, admitted, bedded down, consultations with specialist) are also collected 5 All information is entered into an electronic or manual UMS log 6 Monthly, the senior medical technician/nurse compiles a count (frequency) of each code and sends the information electronically in Microsoft Excel format through the reporting structure (see Fig. 1) single diagnosis could be categorized under more than one code (e.g., the common cold with a cough under no. 7 ENT and no. 8 Respiratory diseases). They also reported that EPINATO lacked specific codes for important clinical problems such as urological, neurological, and cardiovascular disorders. Survey participants reported no previous experience or formal training with the EPINATO system. Other than the Rules of the Game document, no reference manuals or resources were available. Supervisors or colleagues were often consulted to clarify coding classifications even though they also had not received any formal training. This method resulted in coding by judgement call that varied between coders and UMSs. Timeliness At the time of the evaluation, data were coded daily in the UMS and reported monthly through the NATO reporting structure. EPINATO monthly reporting to Deputy Chief of Staff Medical Operations (CF H Svcs HQ) had ceased before the evaluation. Acceptability When stakeholders were asked about their willingness to participate in the system and to use the EPINATO data, many were unsure how health surveillance information could assist them in their current workplace. They also assumed that other military personnel were likely using the data. Stakeholders who felt that health surveillance was valuable expressed concerns about the poor data quality and questioned the system s ability to detect outbreaks and changing disease trends. Representativeness EPINATO data were found to be fairly representative of health events in CF members at the unit and deployed CF population levels although 16% of visits in the chart audit were not coded. EPINATO data specific to CF members working in multinational camps were sent to the lead nation and were not forwarded to Canada for collation. Although the number of CF members working in multinational camps was small, overall morbidity estimates would be underestimated for the total deployed CF population in TFBH. Discussion 959 Given the dynamic nature of the military environment, an effective, valid, and reliable health surveillance system is essential to ensure optimal health of CF members. EPINATO is the standard data collection method for health surveillance for NATO operations. However, there are deficiencies in both the design and the implementation of this tool that limit its usefulness for disease surveillance. The most striking finding in this evaluation was the poor coding reliability for the majority of EPINATO categories which was observed in both the chart audit and SPR studies. Although large outbreaks of disease may still be detected, misclassification in coding can limit early disease or injury detection which is a significant limitation for deployed military operations. Unreliable data likely resulted from multiple factors including: an ambiguous coding tool, a lack of training, and the inherent difficulty in classifying patient presentations (symptoms) in primary care. EPINATO appears to be a simple system with only 25 coding categories. However, this tool requires medical unit staff to understand the overall structure of the classification system and the numerous specific International Classification of Disease codes within each stratum. Studies conducted in civilian primary care settings have shown that coding reliability is poor when physicians and other health professionals are required to understand the overall structure of the system. 7,8 Even with training, the maximum reliability of coding in these studies was approximately 40%. This evaluation also found low user acceptability of EPINATO as a health surveillance tool. The dissatisfaction, for the most part may be related to the poor implementation of EPINATO. Lack of documentation, training, and feedback contributed to a cynical attitude toward data collection without a purpose. Achieving a balance between sensitivity and simplicity is not an easy task but is considered a key attribute in the development of EPINATO. Enhancing training, resources, and feedback to users as well as the automation of coding might lead to improved detection and management of health-related events. Achieving balance between sensitivity and simplicity was a key attribute in the development of EPINATO, which might be better achieved by an automated, electronic surveillance system. Jefferson et al. 1 reported higher coding agreement (67.9%) in their EPINATO evaluation. However, this difference was anticipated given that coding was conducted under optimal conditions. Physicians in their study were aware of the study and underwent training before assigning codes. A strength of the CF evaluation was that we examined the effectiveness of EPINATO in a typical deployment setting. EPINATO codes were assigned by UMS staff before the August 2003 evaluation and the findings reflect actual field performance during a CF operation. Since this evaluation examined EPINATO during one Roto, in one theater of operation, our findings may not be generalizable to other Rotos, nations, or deployments. Another limitation was the nonrandomized selection of clinic visits for the chart audit. Although a random sample of 180 visits was initially planned, this

6 960 CF EPINATO Evaluation in Bosnia-Herzegovina Fig. 1. EPINATO data flow/reporting structure. was not logistically possible and a convenience sample of 128 visits was used instead. Given the availability of the electronic SPR for Roto 12, a second reliability analysis was completed for diseaserelated visits to the UMS. Since the SPR was a census of visits, this data source was considered to be free from selection bias. Also, the larger number of visits in the SPR allowed for weighted values to be completed for each disease-related code. The findings of the chart review and SPR coding reliability studies indicated similar trends with poor agreement for most EPI NATO categories except for eye disorders, ENT, and lower respiratory disorders. Conclusion In the current military environment, limited budgets require justification for resource allocation and health program and policy development must be evidence based. This evaluation found that EPINATO as used by the CF in TFBH was not functioning as an effective health surveillance system. Data quality was poor, user acceptance was low, and subsequently, information was not being analyzed or interpreted to guide public health action. Dedicated resources are required to develop an efficient and effective health surveillance system that can reliably detect disease outbreaks and provide measures of morbidity and mortality for military populations. D FHP is presently investigating the feasibility and design of an automated electronic coding system for deployed Canadian Forces. Electronic International Classification of Disease coding of patient visits would improve the granularity of reports for public health personnel and may increase reliability by automating diagnosis classification. Acknowledgments We thank Cdr I. Fleming, stakeholders, and unit medical staff for their assistance with this evaluation.

7 CF EPINATO Evaluation in Bosnia-Herzegovina References 1. Jefferson TO, Demicheli V, MacMillan A: A pilot study of the introduction of the J95 health data collection system. J R Army Med Corps 1996; 142: Owen JP, Macmillan A: A pilot study of J95 in secondary care in Bosnia. J R Army Med Corps 1998; 144: Centers for Disease Control and Prevention: Guidelines for Evaluating Public Health Surveillance Systems. MMWR Morb Mortal Wkly Rep 2001; 50: Kelsey JL, Whittemore AS, Evans AS, Thompson WD: Methods in observational epidemiology, p 345. New York, Oxford University Press, Inc., Fletcher RH, Fletcher SW, Wagner EH: Clinical Epidemiology: The Essentials, pp Baltimore, MD, Williams and Wilkins, Rosner B: Fundamentals of Biostatistics, Ed 4, p 426. Belmont, CA, Duxbury Press, Gray D, Ward A, Underwood P, Fatovich B, Winkler R: Morbidity coding in general practice. Fam Pract 1989; 6: Nilsson G, Petersson H, Ahfeldt H, Strender LE: Evaluation of three Swedish ICD-10 primary care versions: reliability and ease of use in diagnostic coding. Methods Inf Med 2000; 39: