Author + information
- Received June 22, 1998
- Revision received January 27, 1999
- Accepted February 8, 1999
- Published online June 1, 1999.
- Nathan R Every, MD, MPH, FACC∗,†,* (, )
- Paul D Frederick, MPH, MBA†,
- Michael Robinson, MD‡,
- Jonathan Sugarman, MD, MPH§,
- Laura Bowlby, RN, MBA∥ and
- Hal V Barron, MD, FACC∥,¶
- ↵*Reprint requests and correspondence: Dr. Nathan R. Every, Cardiovascular Outcomes Research Center, 1910 Fairview Avenue E, #205, Seattle, Washington 98102-3620
This study was performed to evaluate whether or not the simpler case identification and data abstraction processes used in National Registry of Myocardial Infarction two (NRMI 2) are comparable with the more rigorous processes utilized in the Cooperative Cardiovascular Project (CCP).
The increased demand for quality of care and outcomes data in hospitalized patients has resulted in a proliferation of databases of varying quality. For patients admitted with myocardial infarction, there are two national databases that attempt to capture critical process and outcome data using different case identification and abstraction processes.
We compared case ascertainment and data elements collected in Medicare-eligible patients included in the industry-sponsored NRMI 2 with Medicare enrollees included in the Health Care Financing Administration-sponsored CCP who were admitted during identical enrollment periods. Internal and external validity of NRMI 2 was defined using the CCP as the “gold standard.”
Demographic and procedure use data obtained independently in each database were nearly identical. There was a tendency for NRMI 2 to identify past medical histories such as prior infarct (29% vs. 31%, p < 0.001) or heart failure (21% vs. 25%, p < 0.001) less frequently than the CCP. Hospital mortality was calculated to be higher in NRMI 2 (19.7% vs. 18.1%, p < 0.001) due mostly to the inclusion of noninsured patients 65 years and older in NRMI 2.
We conclude that the simpler case ascertainment and data collection strategies employed by NRMI 2 result in process and outcome measures that are comparable to the more rigorous methods utilized by the CCP. Outcomes that are more difficult to measure from retrospective chart review such as stroke and recurrent myocardial infarction must be interpreted cautiously.
There is an increased demand by organizations such as payors, healthcare systems, accreditation agencies as well as the public for data on patient outcomes (1–5). Many of these organizations have launched initiatives to collect patient-level process and outcome data. In patients with cardiovascular disease, database research has resulted in effectiveness evaluations of drugs (6), medical procedures (6,7)and processes of care (8–10). Examples of cardiovascular databases range from administrative datasets such as Medicare’s MEDPAR database (11–15), which provides diagnoses, treatments and hospital mortality to clinically rich databases collected as part of randomized trials (16–18). Between these extremes, there are a large number of database models of various scope and quality. The quality of the methods utilized for data collection and analysis is becoming increasingly important because external review organizations as well as the public use these data to evaluate quality of care. Public release of results obtained from studies utilizing Medicare claims data has resulted in both hospital and practitioner identification as providing worse than expected outcomes (19–22).
Patients presenting with acute myocardial infarction (AMI) have been widely studied in various databases due to the large number of cases, the public health threat and the availability of a variety of effective treatments. Although there are many administrative and regional databases that include AMI patients, there are only two with large national samples collected outside clinical trials: the Health Care Financing Administrative (HCFA) database, the Cooperative Cardiovascular Project (CCP) (23)and the pharmaceutical industry-sponsored National Registry of Myocardial Infarction (NRMI 2) (24–26).
These two large datasets attempted to capture information on the same patient populations in many of the same hospitals using markedly different case ascertainment and data collection strategies. The CCP utilized a centralized case identification and internally validated abstraction strategy, whereas NRMI 2 utilizes a local strategy with simpler data collection forms. Although the centralized strategy should result in better case ascertainment and potentially more accurate data collection, the NRMI 2 methods have the advantage of being less expensive, which facilitates ongoing data collection with rapid feedback of results to practitioners. The purpose of the present study was to compare these two datasets to evaluate whether or not the two strategies resulted in similar findings in AMI patients. Specifically, we asked the study question whether or not the simpler case identification and data abstraction processes used by NRMI 2 were comparable with the more rigorous processes utilized in the CCP.
The National Registry of Myocardial Infarction 2 is a multi-center voluntary database designed to collect, analyze and report cross-sectional data on patients admitted with myocardial infarction throughout the U.S. (24). The purpose of NRMI 2 is to provide participating hospitals with periodic summary data assisting local continuous quality assessment activities, to aggregate and test hypotheses describing AMI populations and to monitor the safety experience of a particular thrombolytic agent, alteplase. Between June 1994 and January 1997, 1,529 hospitals enrolled 446,970 patients into NRMI 2.
Data collection in NRMI 2 depends on local patient identification and chart abstraction in combination with centralized data entry. A study coordinator at each hospital is instructed to enroll consecutive confirmed infarcts utilizing local AMI criteria, which commonly includes ECG, cardiac enzyme or angiographic abnormalities or a discharge diagnosis of 410.xx electrocardiogram (acute myocardial infarction by ICD coding). Coordinators attend a half-day training course and are provided with a reference manual that includes case report form field definitions and examples of correct responses. Patient information is transcribed onto a two-page case report form. Completed forms are forwarded to an independent central data collection center, ClinTrials Research, Inc. (Lexington, Kentucky). There are 87 electronic data checks to detect internal inconsistencies, omissions and out-of-range values. Case report forms that fail edit checks are mailed back to the study coordinator for data resolution.
The Cooperative Cardiovascular Project database is a national effort by the HCFA to improve care for Medicare beneficiaries discharged with AMI. Unlike NRMI 2, the CCP is a mandated quality improvement process and is therefore included in the universe of hospitals providing care for Medicare beneficiaries with AMI.
Patients with the discharge diagnosis of AMI (410.xx, excluding 410.x2 [re-admission for AMI within eight weeks of discharge]) were retrospectively identified by HCFA utilizing a central administrative database (23). All hospitals that admitted patients with Medicare insurance were then required to submit the photocopied medical records of the identified AMI patients to a central data abstraction center. Acute-care hospitals from across the country (n = 4,223) submitted 224,377 medical records for AMI patients discharged within specified eight-month periods between February, 1994 and July, 1995. Clinical Data Abstraction Centers (CDACs) performed data abstraction from medical records.
All CDAC abstractors receive substantial training in applicable medical terms and procedures. In addition, CDACs periodically perform blinded reabstractions on random samples of medical records to measure the quality of the abstraction process. Thus, in comparison with NRMI 2, the CCP is a database that utilized centralized case identification and data abstraction.
Internal validity was evaluated at the hospital and patient level. To evaluate case ascertainment and overall comparability of the databases independent of individual patient matching, we performed a hospital-level comparison. In this comparison we attempted to evaluate whether or not differences in patient identification procedures would result in differences in baseline demographic, patient presentation, process of care and outcome calculated in each database. For comparison purposes, the CCP was designated to be the reference database.
Because there were hospitals in NRMI 2 not included in the CCP and vice versa, hospitals were first matched based on Medicare provider number. Patients discharged from hospitals not participating in both databases were excluded. Next, a set of discharge dates common to both databases was defined for each hospital and patients discharged outside these inclusive dates were excluded. To define a patient population that should have been Medicare-eligible, we excluded patients 64 years or younger. Finally, because NRMI 2 does not include unique patient identifiers that allow linking medical records in those patients who were transferred, patients who were either transferred in or transferred out of the index hospital were excluded from the analysis. Using this process, we identified 35,675 patients in NRMI 2 and 42,703 patients in the CCP that were admitted to 1,087 matching hospitals during identical enrollment periods.
To evaluate internal validity at the individual patient level, we performed a patient-level analysis. In this analysis, individual patients enrolled in both databases were matched, and common variables collected in both databases were compared. Patient matching was based on composite keys that were developed based on shared attributes within the data schema. The composite keys were classified into a 4-key (hospital, discharge date, age and gender), 5-key (4-key plus arrival date) and 6-key (5-key plus arrival time). To match records among the CCP and NRMI 2 comparison populations, an iterative program module was developed to identify 6-key, 5-key and 4-key matches. When all attributes within the key were the same between NRMI 2 and CCP comparison populations, the match was considered exact. Discharge and arrival date were permitted an allowance of ±1 day, age ±1 year and arrival time ±1 h. The highest priority patient match occurred at the 6-key level.
The preliminary patient-level database contained 28,689 records in which 24,908 (87%) were exact matches. Among the exact matches, 14,531 (58%) were based on the 6-key match, 9,940 (40%) on 5-key and 437 (2%) on 4-key. There were also 3,025 duplicate cases removed from the preliminary match resulting in a final matched NRMI 2/CCP comparison population of 25,664 patient episodes contributed by 1,076 hospitals.
External validity was evaluated by comparing baseline data, process and outcome of care in all patients 65 years and older in NRMI 2 and CCP without matching at either the patient or hospital-level (unmatched comparison). For this comparison, there were 73,774 patients admitted to 1,338 hospitals that participated in NRMI 2 and 129,482 patients admitted to 4,205 hospitals in the CCP.
In the hospital-level and unmatched comparison, baseline demographics, clinical events, use of medications, utilization of cardiac procedures and outcomes were compared using chi-square for categorical and Student ttest for continuous variable comparisons. Variables chosen for comparison were those included in both databases and were predefined based on previous studies evaluating outcome in AMI patients. Logistic regression was utilized to determine the multivariate association between the variables and in-hospital mortality in each database to compare whether the calculated odds ratios were different depending on the database. All statistical calculations were performed with the SAS 6.12 statistical procedure (SAS Institute).
In the patient-level comparison, percent agreement and Kappa scores were calculated for variables collected in each dataset. Percent of agreement was defined as the proportion of agreement plus the proportion of those who did not agree divided by the total number of matched patients.
There were a total of 446,970 patients included in the NRMI 2 database and 224,377 in the CCP database. From these sources, we identified 35,675 AMI patients in NRMI 2 and 42,703 patients in CCP who were likely to be Medicare-eligible and who were admitted to the same hospitals during the same data abstraction period. From these two groups, we were able to match 25,664 patients that represented identical patients enrolled in both databases and are referred to as the patient-level comparison. Patients included in the CCP but not identified in NRMI 2 (CCP unmatched) (Table 1)were similar to the entire CCP cohort. There were no substantial differences in demographics, past medical histories or hospital course. Hospital mortality was slightly higher in CCP patients not identified in NRMI 2 (18.7% vs. 18.1%, p = 0.08). Patients included in NRMI 2 but not identified in the CCP (NRMI 2 unmatched) were similar to the entire NRMI 2 cohort in terms of demographics and past medical history. However, hospital complications including shock (8.3% vs. 7.8%, p = 0.09), stroke (2.3% vs. 1.9%, p = 0.008) and hospital mortality (24.1% vs. 19.7%, p = 0.001) were each higher in the NRMI 2 patients not identified in the CCP.
In the first analysis, we compared selected variables reported in hospitals included in both databases during matched time periods (hospital-level comparison in which there was no individual patient matching). Baseline demographic data and past medical histories were similar in the two databases (Table 2). Although there were statistically significant differences in most comparisons due to the large numbers of patients in each cohort, there was little difference in age, gender or race. In general, the CCP database was more likely to identify past diseases such as prior myocardial infarction or heart failure. There was little difference in the identification of past cardiac procedures.
Process of care variables were identified with similar frequency in each database (Table 3). National Registry of Myocardial Infarction 2 was more likely to identify the use of thrombolytic therapy in AMI patients (15.6% vs. 14.6%, p < 0.001), but there was little difference in the identification of the use of cardiac catheterization, bypass surgery or coronary angioplasty. Identification of discharge medications was similar in both databases, although the CCP was somewhat more likely to identify the use of aspirin, angiotension-converting enzyme inhibitors and beta-adrenergic blocking agents at hospital discharge. The mean length-of-hospital stay was nearly identical.
After hospital admission, NRMI 2 was less likely to identify hospital complications such as shock (7.8% vs. 8.3%, p = 0.025) or stroke (1.9% vs. 3.4%, p < 0.001) (Table 3). Most stroke identification in NRMI 2 was in patients with intracranial bleeding (1.6%); thus, there appeared to be underidentification of nonhemorrhagic strokes in NRMI 2. Hospital mortality was higher in NRMI 2 (19.7% vs. 18.1%, p < 0.001). This may be explained by the inclusion of noninsured elderly patients (although within the Medicare age group eligibility) in NRMI 2 that appeared to have higher mortality than those identified with Medicare insurance (24.5% vs. 19.9%, p = 0.06).
To further illustrate the comparability of the two datasets using the hospital-level comparison (e.g., without individual patient matching) we constructed two logistic regression models that evaluated the association between variables common to both datasets and hospital death. Each model was run first in the CCP cohort and then in the NRMI 2 cohort. Using a set of predefined variables, each database model predicted a similar association between the variable and hospital death (Fig. 1). For example, the association between age and hospital death was (odds ratio = 1.18 vs. 1.25 per year for NRMI 2 and the CCP model, respectively). There was less correlation with the variables of reinfarction and shock (odds ratio = 5.90 vs. 2.98 [reinfarction] and odds ratio = 33.85 vs. 23.18 [shock] for NRMI 2 and the CCP model, respectively). In a second model, we added variables included in CCP that were not abstracted in NRMI 2. In this expanded model, the CCP variables terminal illness (odds ratio = 1.6, 95%, confidence interval = 1.1–2.1) and apache 2 score (odds ratio = 1.12, 95%, confidence interval = 1.11–1.13) were significantly associated with hospital mortality. The history of dementia or COPD were not associated with hospital mortality. The odds ratios illustrated in Figure 1did not substantially change when these additional variables were added to the model.
To compare data quality at the patient level, we performed an analysis that compared individual variables collected in both databases on 25,664 matched patients. Because patients were matched by both age and gender, there was nearly 100% agreement on these variables (Table 4). However, there was also good agreement in most other demographic and historic variables collected in both databases (kappa values = 0.61–0.99). Consistent with the hospital-level comparison, prior history variables such as prior infarct (88% agreement) or heart failure (87% agreement) were somewhat underascertained in NRMI 2. Process of care variables such as procedure use (cardiac catheterization—98%, bypass surgery—100% and coronary angioplasty—98%), thrombolytic therapy use (98%) and discharge medications showed high levels of agreement (Table 5, kappa values = 0.75–0.97). There was close agreement on the hospital mortality end point (kappa = 0.98). One-hundred ten patients were classified as hospital deaths in NRMI 2 but classified as alive in the CCP, whereas 60 patients were classified as hospital deaths in the CCP but classified as alive in NRMI 2 (Table 6). Of clinical end points, identification of reinfarction had the lowest kappa value (0.21) with disagreement in 1,222 of 25,664 patients.
Because the CCP included all non-Federal hospitals (n = 4,223) while NRMI 2 (n = 1,338) included select hospitals, we compared Medicare eligible patients in each database without matching at either the hospital or patient level to evaluate external validity. Hospitals participating in NRMI 2 were larger on average (264 vs. 174 beds, p = 0.0001), more likely to have on-site catheterization facilities (72.6% vs. 42.9%, p = 0.001) and on-site bypass surgery (39.3 vs. 20.3, p = 0.001). Despite these differences in hospital characteristics, most demographic, process and outcome data were similar (Table 7). As with the other analyses, past histories were modestly underreported in NRMI 2 with the greatest difference noted in past history of stroke (12.0% in NRMI 2 vs. 15.5% in CCP, p = 0.001). Patients included in NRMI 2 were more likely to undergo thrombolysis, cardiac catheterization, coronary angioplasty or bypass surgery. Clinical event reporting was similar except for underreporting of stroke (2% vs. 3.5%, p = 0.001) and heart failure complication (24.7% vs. 47.8%, p = 0.001) in NRMI 2. Hospital mortality was similar in the two databases.
The demand for process and outcome measures in American healthcare systems is growing. The traditional mechanism for obtaining these measures has been the administrative dataset, which has the advantage of identifying all patients with a particular diagnosis or procedure (11–15). The disadvantage of administrative data is the lack of clinical detail (27). At the other extreme are databases that are obtained in the setting of clinical trials which have the advantage of clinical detail and accuracy but are resource-intensive and limited to the select patient population that qualifies for the trial (16–18). The challenge for physicians, researchers and administrators interested in measuring process of care and outcome is to develop efficient mechanisms to identify patients and collect data that are both internally and externally valid.
In this study, we compared two different mechanisms to collect process and outcome data in patients with AMI. The CCP database relies on a centralized system of patient identification and data abstraction that assures accuracy through standardized data abstraction procedures and random reabstraction methods. Although this methodology should be considered the “gold standard,” it is time intensive and therefore expensive and does not lend itself to continuous data collection. The NRMI 2 database, on the other hand, relies on local hospital patient identification and chart abstraction without methods for assuring consecutive patient enrollment or chart reabstraction for purposes of validation. This data collection process is less time intensive and therefore can be used as an ongoing registry. However, the less rigorous data collection methodology used in NRMI 2 has led some to question the validity of findings. For example, if hospitals excluded patients with poor outcomes or processes of care, this would bias overall results reported from the NRMI 2 database.
Our findings, however, confirmed few differences in these two datasets despite the substantial difference in data collection methodologies. Although there was a modest underascertainment of potentially eligible infarct cases (10%–15%), NRMI 2 was comparable with the CCP in defining demographic data, past medical histories and procedure and medication use. Surprisingly, overall mortality in the hospital-level comparison was somewhat higher in NRMI 2 than the CCP database. We had hypothesized that NRMI 2 investigators may underreport poor outcomes, but this did not seem to be the case. It appears that the higher mortality calculated in NRMI 2 patients was due to inclusion of non-Medicare insured patients in NRMI 2 which would not have been identified in the CCP. In fact, in 25,664 matched patients, there was disagreement in the mortality end point in only 0.2%.
Of some concern is a higher level of disagreement (lower kappa values) at the patient-level analysis in other clinical end points such as shock, stroke or recurrent infarction. One explanation for these findings could be different variable definitions used in each database. For example, NRMI 2 defines recurrent myocardial infarction as “the occurrence of another MI confirmed by new diagnostic ST/T wave changes or a second elevation of cardiac enzymes >2 times normal.” The CCP, on the other hand, defines the same variable as “a new event during the hospitalization in which more heart muscle is damaged.” Clinicians may also use different definitions for these events that may lead to variation in chart documentation. Because of the lower kappa values, these outcomes (shock, stroke and recurrent infarction) must be viewed with caution.
The external validity of NRMI 2 was also reasonable as compared with the CCP. Data obtained from NRMI 2 would underidentify comorbidities and overidentify cardiac procedure use as compared with the population as a whole. These differences were quite small and probably not clinically meaningful. Although mortality outcomes were similar, the rate of stroke and heart failure during the hospitalization was substantially different between databases.
Although not tested in this study, we might speculate why the results obtained from NRMI 2 were comparable with the more rigorous CCP database. First, identification of AMI patients is relatively simple, because nearly all patients are treated in the hospital, most are admitted to one area of the hospital (CCU) and there is a financial incentive to code with the AMI discharge diagnosis (410). Second, experience with NRMI 1 gave the study sponsors and advisors valuable information on the design and use of the NRMI 2 data collection form (24,28). This allowed the data collection form to include most critical data while being efficient and “user-friendly” to complete. Third, many hospitals participating in NRMI 2 use the quarterly report describing individual hospital results in comparison with National standards, as an internal quality assurance mechanism. This may promote careful patient identification and chart abstraction in those institutions. Finally, hospitals receive a modest payment for each case report form that is completed, which may promote participation by institutions with limited resources.
Although these two independently collected databases were comparable, there are several limitations of the study. First, AMI is probably a simpler disease to study in a registry such as NRMI 2 due to the hospitalization of most patients and readily measured process and outcome variables. Our findings are not easily generalized to other diseases. Second, the incentives for quality data collection in NRMI 2, such as the use of quarterly reports for quality improvement projects, are likely critical components of the validity of the Registry. Our findings are probably not easily generalized to other registries with different procedures and incentives to assure data quality. Finally, although we assumed that costs for NRMI 2 were lower than CCP based on time of chart abstraction and the absence of chart copying and mailing to a central data collection center, we were unable to measure costs directly.
We also compared two databases that relied on retrospective chart abstraction to define variables. Although we defined the CCP as the “gold standard,” chart abstraction may not be the optimal method for defining process and outcome of care. For some healthcare systems, electronic medical records with point of care prospective data collection using standardized data definitions have replaced paper charts and are considered by some to be the optimal data collection method. Although we could not compare either the CCP or NRMI 2 with an electronic medical record, the paper chart is still used in the majority of U.S. hospitals and thus represents the best available record of process and outcome of care.
The findings from this study have important policy implications regarding national and regional efforts to assess and improve care among patients with AMI. The CCP, which had the advantage of including the universe of Medicare beneficiaries discharged from hospitals in a specific time period, required a substantially greater commitment of resources than the approach taken with NRMI 2. That is, hundreds of thousands of medical records were photocopied and submitted to a central site, and medical records were abstracted by reviewers not familiar with each facility’s records and conventions. Hospital records of non-Medicare beneficiaries were not included. In addition, considerable time—often two years or more—elapsed between the delivery of care and reporting of the data back to hospitals, and only aggregate data are reported.
In contrast, the NRMI 2 data were efficiently abstracted from the original medical records, returned to hospitals on a quarterly basis with local and national comparisons and included patients discharged with AMI regardless of payment source. The timeliness, continuous feedback and involvement of local care providers in NRMI 2 data collection and analysis are all characteristics recognized to facilitate quality improvement activities.
The efficiencies gained through the NRMI 2 data collection process result in some limitations of the Registry. First is the underascertainment of eligible patients with AMI. It is unlikely that this underascertainment has a major influence on the use of the Registry as either a research tool or a quality improvement instrument. This is because AMI patients excluded from NRMI 2 appear to be similar to those included in the database and would be an unlikely source of bias. A second limitation of NRMI 2 is disagreement at the individual patient level about the occurrence of clinical events such as stroke or recurrent infarction. Because we did not observe substantial differences in this variable at the hospital-level analyses, it is unlikely that misclassification would influence research results based on NRMI 2 data. As a quality assurance tool, misclassification of sentinel events in NRMI 2 could be problematic. However, because hospitals collect and interpret their own data, validation of data at the patient level is under direct control of those receiving the quality reports. For institutions using NRMI 2 as an internal quality assurance process, it is critical to insure adequate training of local abstractors. Institutions might also consider random reabstraction of charts to insure data reliability. Finally, the local case identification and chart abstraction processes along with patient confidentiality issues do not allow NRMI 2 to determine outcome in patients after hospital transfer and after discharge. Partnerships with other administrative datasets may allow these linkages to take place in the future.
We conclude that the simpler case ascertainment and data collection strategies employed by NRMI 2 result in process and outcome measures that are comparable with the more rigorous methods utilized by the CCP. Outcomes that are more difficult to measure from retrospective chart review such as stroke and recurrent myocardial infarction must be interpreted cautiously.
☆ This study was supported by grant 94-304, VHA Health Services Research Career Development Award and grant 638309 from the Health Care Financing Administration and was developed as a part of a quality improvement project conducted under the Health Care Quality Improvement Program of the Health Care Financing Administration, Baltimore, Maryland. The conclusions and opinions expressed and the methods used are those of the authors and not necessarily the policy of the Health Care Financing Administration. The authors assume all responsibility for accuracy and completeness.
Performed as a Quality Assurance Project in conjunction with ProWest and the Healthcare Financing Administration.
- acute myocardial infarction
- Cooperative Cardiovascular Project
- Clinical Data Abstraction Centers
- Health Care Financing Administration
- NRMI 2
- National Registry of Myocardial Infarction
- Received June 22, 1998.
- Revision received January 27, 1999.
- Accepted February 8, 1999.
- American College of Cardiology
- MITI Investigators,
- Every N.R.,
- Parsons L.P.,
- Hlatky M.,
- Martin J.M.,
- Weaver W.D.
- Weaver W.D.,
- Parsons L.,
- Every N.R.
- Jollis J.G.,
- Peterson E.D.,
- Nelson C.L.,
- et al.
- Pilote L.,
- Califf R.M.,
- Sapp S.,
- et al.
- Harrington R.A.,
- Lincoff A.M.,
- Califf R.M.,
- et al.
- ↵Pennsylvania Healthcare Costs Containment Council: A Consumer Guide to Coronary ARTERY Bypass Graft Surgery, Volume IV:1993 Data. Harrisburg (PA): Pennsylvania Health Care Cost Containment Council, 1995.
- Wilson P.,
- Smoley S.R.,
- Werdegar D.
- Rogers W.J.,
- Bowlby L.J.,
- Chandra N.C.,
- et al.
- Becker R.C.,
- Gore J.M.,
- Lambrew C.,
- et al.
- Canto J.G.,
- Rogers W.J.,
- Bowlby L.J.,
- French W.J.,
- Pearce D.J.,
- Weaver W.D.
- Maynard C.,
- Weaver W.D.,
- Lambrew C.,
- Bowlby L.J.,
- Rogers W.J.,
- Rubison R.M.