Search

One common explanation for the failure to achieve broad physician utilization of computer applications has been the suggestion that "We can't teach older individuals to use computers." To investigate this hypothesis, we examined utilization patterns for the Clinical Information System (CIS) at Columbia Presbyterian Medical Center (CPMC). We analyzed CIS usage for 925 attending physicians who were listed as an admitting or attending physician or surgeon for at least one patient during the year 1992. Sixty-one percent (561/925) of the attending physicians used the system at least once during the year. Sixty five percent (186/287) of the physicians who admitted at least 50 cases used the system at least 120 times during the year. The most surprising aspect of our analysis was that physicians in their late 60's and early seventies actually used the system more than their peers who were in their late 50's. Patterns of use by age group were similar for those who admitted many and few patients to the hospital. Using linear regression and chi squared analysis, we found that age is correlated (p < 0.002) with levels of physician use (inquiries per case), although age can explain (r-squared) only 3% of the observed variation in utilization patterns. We also found that there was significant variation in utilization (inquiries per case) by attendings in different departments (p < 0.007). However, the variation within departments was also large. We conclude that age and type of practice are statistically significant but not major factors in predicting which attendings will use the system. Growth rates over time (19% year to year increase in the average number of different users per day) indicate that, if present trends continue, virtually all physicians regardless of age will use the Clinical Information System for results review. We continue to feel that providing value, access and ease of use are the most important determinants for success.

ASTM subcommittee E31.15 on Health Knowledge Representation was formed to promote standards for defining and sharing health knowledge bases. Its first standard, the Ardan Syntax, is focused on knowledge bases that can be represented as a set of independent modules called Medical Logic Modules (MLMs). The standard is in clinical use and has generated significant interest in industry and academics. The Extensions task group plans to extend the syntax where appropriate, to expand to other types of knowledge bases. The Validation/Verification task group is approaching the enormous problem of evaluating knowledge bases and the process of sharing them.

Linking a knowledge-based system (KBS) to a clinical database is a difficult task, but critical if such systems are to achieve widespread use. The Columbia-Presbyterian Medical Center's clinical event monitor provides alerts, interpretations, research screening, and quality assurance functions for the center. Its knowledge base consists of Arden Syntax Medical Logic Modules (MLMs). The knowledge base was analyzed in order to quantify the use and impact of KBS-database links. The MLM data slot, which contains the definition of these links, had almost as many statements (5.8 vs. 8.8, ns with p = 0.15) and more tokens (122 vs. 76, p = 0.037) than the logic slot, which contains the actual medical knowledge. The data slot underwent about twice as many modifications over time as the logic slot (3.0 vs. 1.6 modifications/version, p = 0.010). Database queries and updates accounted for 97.2% of the MLM's total elapsed execution time. Thus, KBS-database links consume substantial resources in an MLM knowledge base, in terms of coding, maintenance, and performance.

The concept of "one-stop information shopping" is becoming a reality at Columbia-Presbyterian Medical Center. Our goal is to provide access from a single workstation to clinical, research, and library resources; university and hospital administrative systems; and utility functions such as word processing and mail. We have created new organizational units and installed a network of workstations that can access a variety of resources and systems on any of seventy-two different host computers/servers. In November 1991, 2,600 different individuals used the clinical information system, 700 different individuals used the library resources, and 900 different individuals used hospital administrative systems via the network. Over the past four years, our efforts have cost the equivalent of $23 million or approximately 0.5% of the total medical center budget. Even small improvements in productivity and in the quality of work of individuals who use the system could justify these expenditures. The challenges we still face include the provision of additional easy-to-use applications and development of equitable methods for financial support.

Over the past seven years, Columbia-Presbyterian Medical Center has been planning and implementing an integrated academic information management system. Accomplishments to date include establishing an institutional information architecture, installing a campus-wide network of workstations, recruiting the staff needed to develop and implement the system, and developing various applications. This paper presents the rationale and steps involved in these accomplishments, as well as data on use of the system so far.

We have developed a computerized method for performing bibliographic searches directly from patient data involving five steps: 1) identifying specific patient data which raises a question in the mind of the user, 2) selection (from a list of generic questions) of a small number of questions which fit the selected patient data, 3) automated translation of the patient data into appropriate terms used for bibliographic indexing, 4) conversion of the question selected by the user into a search strategy, and 5) transfer of the search strategy to a search engine for a bibliographic database. We have modified the Columbia-Presbyterian Clinical Information System to experiment with this method. The first implementation converts patient diagnoses and procedures coded in ICD9-CM into Medical Subject Headings (MeSH) and searches Medline using BRS/Onsite. Challenges include development of a useful set of generic questions and translation from ICD9-CM to MeSH using the Unified Medical Language System (UMLS).

The developing Integrated Academic Information System (IAIMS) at Columbia-Presbyterian Medical Center provides data sharing links between two separate corporate entities, namely Columbia University Medical School and The Presbyterian Hospital, using a network-based architecture. Multiple database servers with heterogeneous user authentication protocols are linked to this network. "One-stop information shopping" implies one log-on procedure per session, not separate log-on and log-off procedures for each server or application used during a session. These circumstances provide challenges at the policy and technical levels to data security at the network level and insuring smooth information access for end users of these network-based services. Five activities being conducted as part of our security project are described: (1) policy development; (2) an authentication server for the network; (3) Kerberos as a tool for providing mutual authentication, encryption, and time stamping of authentication messages; (4) a prototype interface using Kerberos services to authenticate users accessing a network database server; and (5) a Kerberized electronic signature.

In a radiology department, clinical audit implies multiple readings of selected images to identify those findings that should be recognized and to document any departure from this standard for each radiologist. The authors developed an alternate approach for an audit on the basis of clinical outcomes collected in a medical computing facility. Techniques borrowed from information theory were used to measure the clinical information contributed by radiologists as they interpreted chest radiographs. The reported findings were evaluated in light of the discharge diagnosis. The scores generated quantified the information contributed to the final diagnosis by the radiologist's description. This audit approach was tested in a group of 100 chest radiographs. Significant differences were found in the mean scores for information contributed by five different readers. These differences were similar to differences demonstrated in audits by means of multiple readings of chest radiographs. These results support use of a form of audit that is substantially less expensive and time consuming than that typically used in radiology departments.

Columbia-Presbyterian Medical Center is implementing a decision-support system based on the Arden Syntax for Medical Logic Modules (MLM's). The system uses a compiler-interpreter pair. MLM's are first compiled into pseudo-codes, which are instructions for a virtual machine. The MLM's are then executed using an interpreter that emulates the virtual machine. This design has resulted in increased portability, easier debugging and verification, and more compact compiled MLM's. The time spent interpreting the MLM pseudo-codes has been found to be insignificant compared to database accesses. The compiler, which is written using the tools "lex" and "yacc," optimizes MLM's by minimizing the number of database accesses. The interpreter emulates a stack-oriented machine. A phased implementation of the syntax was used to speed the development of the system.

This paper presents methods used to develop a conceptual model for a patient database forming the centerpiece of a clinical information system under development. Various modeling techniques are discussed using a simplified fragment of the model. A method for mapping the model onto a relational design optimized for single patient retrievals is described. The results section discusses a number of issues pertaining to the flexibility and usability of this architecture.

The concept of "one stop information shopping" is becoming a reality at Columbia Presbyterian Medical Center (CPMC). The goal of our effort is to provide access to university and hospital administrative systems as well as clinical and library applications from a single workstation, which also provides utility functions such as word processing and mail. Since June 1987, CPMC has invested the equivalent of $23 million dollars to install a digital communications network that encompasses 18 buildings at seven geographically separate sites and to develop clinical and library applications that are integrated with the existing hospital and university administrative and research computing facilities. During June 1991, 2425 different individuals used the clinical information system, 425 different individuals used the library applications, and 900 different individuals used the hospital administrative applications via network access. If we were to freeze the system in its current state, amortize the development and network installation costs, and add projected maintenance costs for the clinical and library applications, our integrated information system would cost $2.8 million on an annual basis. This cost is 0.3% of the medical center's annual budget. These expenditures could be justified by very small improvements in time savings for personnel and/or decreased length of hospital stay and/or more efficient use of resources. In addition to the direct benefits which we detail, a major benefit is the ease with which additional computer-based applications can be added incrementally at an extremely modest cost.

Computerized medical logic has been developed to calculate the patient-specific pretest likelihood of pleural fluid for radiographic examinations. This medical logic was determined by searching the computerized data base for clinical indicants which are found to differ between groups of patients with and without pleural fluid. By using a priori probabilities of pleural fluid and sequential application of Bayes' equation to revise the likelihood according to the presence of significant indicants, patient-specific likelihood were calculated. This medical logic was tested on a group of 591 patients with and without pleural fluid by radiographic evidence. The results indicated a sensitivity of 95% and a specificity of 81%.

We have developed a computer-administered history designed to directly interview hospitalized patients with pulmonary disease. A frame-based decision system is used to direct the history and to generate a one- to five-member differential diagnostic list based on this history. This system incorporates a cognitive model of question selection and a Bayesian scoring algorithm. Structures to control the choice of questions are embedded in the diagnostic frames and in a QUERY program that makes the final choice of questions. We have compared the behavior of this decision-driven approach with a history taken using a paper questionnaire. The paper-based history presents 182 questions to every patient and captured 75% of 85 pulmonary diseases in its differential lists. The decision-driven system asks 50.7 +/- 31.0 (mean +/- standard deviation) and captured 74% of 61 pulmonary diseases. Our experience suggests that the use of a computerized diagnostic knowledge base to direct the selection of pertinent questions can substantially reduce the number of questions necessary to collect a diagnostically useful patient history.

For a group of 658 patients who received coronary artery bypass graft surgery, we investigated the correlation between the degree of early (6 months) graft patency and recurrence of anginal symptoms, late myocardial infarction, and postoperative coronary-related death. The patients were grouped according to the number of surgically placed grafts, and each group was further subgrouped on the basis of the number of grafts functioning at the early postsurgical follow-up examination. The patients were observed over a period as long as 13 years. The frequency with which angina returned correlated significantly with the degree of patency within each of the groups (one, two, three, or four grafts); patients with a higher percentage of patent grafts experienced longer periods of freedom from angina. On the average, patients with all of their multiple grafts patent experienced at least 7 more years of symptomatic relief than their counterparts with all grafts occluded. Most surprisingly, the rate of the return of angina for those patients who had all grafts patent and were completely revascularized was independent of the number of diseased vessels or the number of grafts placed. The findings for coronary death and postoperative infarction showed similar trends.

An f/100, 12.2-m focal length photoelectric telescope designed specifically for daytime astrometry of objects near the sun is now operative at its Tucson, Arizona, site. The design goal was to achieve accuracies of order 0.001 sec of arc in field position measurements of stars. To accomplish this, many features reducing systematic and random errors are employed, including Schupmann medial telescope optics, compensation for lateral color aberration, apodization for reduction of diffracted light, and use of an accurately measured solar diameter for calibrating the field.

We traced left ventricular contours, frame-by-frame throughout systole and diastole, of normal sinus beats from 30 degrees right anterior oblique ventriculograms from 32 normal patients. We separated both systole and diastole into 19 equal time intervals each and calculated regional lengths (R), normalized by both end-diastole length and relative time interval (T) in systole and diastole and diastole, for the middle inferior wall, distal inferior wall, apex, distal anterior wall, middle anterior wall, and proximal anterior wall. We also computed the relative velocities of R, delta R/ delta T, over each quarter of systole and diastole. Comparing systole with diastole, we found significant differences between paired values of R at all regions except the distal inferior wall, but these differences were not the same between regions. Between regions, mean R and delta R/ delta T values were significantly different as early as the first quarter of systole. Within a region, there were significant differences between mean R and delta R/ delta T values over intervals as short as one-fourth of systole or diastole. Thus, there is no homogeneity between regions in normal wall motion in both systole and diastole. This normal lack of homogeneity has important clinical implications for identifying abnormal wall motion in individual patients from ventriculographic measurements, and for using the information present in the diastole portion of the ventriculogram to characterize normal segmental function.

Initial results obtained with a simple, fully automated algorithm for detection of left ventricular boundaries are presented. The strength of this approach is the use of dynamic programming search techniques, which allow determination of local border points to be influenced by the entire global border location. The relative contributions of mask mode subtraction and the dynamic search technique are evaluated with respect to accurate border definition. These computer-determined ventricular borders are compared with hand-traced borders on subtracted and unsubtracted images. The modular dynamic search algorithm is shown to perform better than previously described algorithms, which generally require operator interaction. It is also shown that for both manual and automated techniques, ventricular borders derived from subtracted images may be significantly different from borders derived from nonsubtracted images.

We evaluated changes in ventricular wall motion after surgery by comparing smoothed, filtered measurements of regional percent shortening (RPS) from right anterior oblique ventriculograms in 37 patients before and after surgery. After surgery there was a significant (p less than 0.05) decrease in the number of regions with hypokinetic wall motion. The distribution of RPS values was also different (p less than 0.005). However, the mean value of RPS for the surgery group as a whole was not significantly altered. These data were contrasted with RPS data from 11 control patients, who were each studied twice but did not have surgical intervention. Similar analysis of the control group did not show any significant change between studies in the number of hypokinetic regions, and the distributions of RPS for the first and second angiograms were not different. We found a 10.3% absolute mean change in repeated measurements of RPS in the control group. We conclude that significant changes occurred after surgery that were not evident in the control group, and the amount of variability in repeated measurements of RPS suggests that analysis should be applied to group rather than individual data.

We designed an automated system for managing large-scale screening and diagnostic mammography. The system collects coded mammographic findings from the radiologist and records a history directly from the patient. This information is stored in an integrated clinical data base to which the results of subsequent examinations or surgery are added. In addition, the system generates letters to the patient and her physician that describe mammographic findings and letters reminding them of routine screening visits. For patients who have positive results on examinations, it checks for records of biopsy or repeat mammography and generates follow-up letters if appropriate intervention is not found. While this system is part of a comprehensive computerized hospital information system, mammography management tools with most of the features described can be designed for relatively inexpensive microcomputers.

A computerized technique for measuring total lung capacity from radiographic chest films based on methods originally suggested by Barnhard and associates has been developed. Total lung capacity is automatically determined by tracing the margins of the lungs in both posteroanterior and lateral radiographs with a commercially available hypersonic coordinate digitizer. The results compare favorably with those obtained using a body plethysmograph. This simple, automatic procedure requires minimal training, is rapid, and can be easily implemented.

The sequence for the onset of segmental contraction of the left ventricle was studied in 25 normal patients by analyzing sequential frames obtained at 16.7-msec intervals of right anterior oblique (RAO) ventriculograms by two independent methods. In the first method, we compared the times of onset of contraction of the hemidiameters associated with each of 54 segments with time of onset of contraction of the average of all the hemidiameters for the ventricular contour. In the second method we used a radial coordinate system and determined relative phase relationships by plotting the motion of each of 54 segments against the average motion of all segments. The resulting pattern showed that, on the average, the midregion of the inferior wall began to contract 25 msec before the apex and the midregion of the anterior wall began contraction 18 msec before the apex. In 12 of 25 patients the interior and anterior walls both began to contrast before the apex. In only one of 25 patients did the apex begin to contract first. This sequences of contraction corresponds to the reported sequence of electrical activation for normal human left ventricles.

Demographic and symptomatic data gathered from 460 patients referred for lumbosacral CT examinations were analyzed to determine if the prescan probability of normal or abnormal findings could be predicted accurately. We were unable to predict the presence of herniated disk on the basis of patient-supplied data alone. Age was the single most significant predictor of an abnormality and was sharply related to degenerative disease and spinal stenosis.

Coronary revascularization has been reported to have failed to effectively rehabilitate working-age patients. This study of 565 patients demonstrates that motivation to return to work is strongly influences by age and educational level. Patients under age 55 are more likely to return to work than are patients over that age, but preoperative job classification does not influence rehabilitation. Although preoperative disability was associated with a slightly lower return-to-work rate (90%) than was the case with patients working preoperatively (97%) preoperative retirement was a strongly negative influence on rehabilitation. In this study, 80% of the patients worked to or beyond retirement age, and duration of work was not influenced by preoperative disability. The salary produced by those patients who were rehabilitated by surgery was four and a half times greater than the total cost of care and disability payments for the entire patient population. The factors which seemed to be the most important in effective rehabilitation were the psychological preparation of patients and their families and the attitude toward rehabilitation expressed by physicians and employers.

In 405 men under the age of 65 requiring coronary bypass, in whom 6 month postoperative arteriograms were performed, we compared graft patency to postoperative work status and recurrent symptoms. We divided the population into subgroups of patients with varying degrees of patency. When these subgroups were tested, no significant dependence was found between the degree of graft patency and the percentages of patients who were working after operation, unless certain subgroups were removed from the population. When relief of angina was examined in the same manner as the postoperative work, we found a significant dependence between the graft patency and the percentage of patients who reported either short or long-term relief of angina in all groups. Because the rate of rehabilitation was high, even in patients with occluded grafts, and because older patients were less likely to return to work than younger patients despite successful revascularization, we conclude that physician emphasis on work rehabilitation and patient age, as well as graft patency, are all important factors which influence postoperative work status.

Optimal visualization of moving structures such as the heart and coronary arteries using digital radiographic imaging systems is a difficult problem that can involve tradeoffs between temporal, spatial, and density resolution. The motion dependence of four angiographic measures of vessel dimensions is given and demonstrated experimentally. Although the densitometric cross-sectional area is shown to be independent of motion, densitometric thickness (contrast) decreases and the apparent width (distance between edges) and densitometric width both increase with motion. Knowledge of the velocity allows the apparent width but not the densitometric width to be corrected. It is also shown that current limitations which result in tradeoffs between kVp and x-ray pulse duration seriously compromise the advantages of using small focal spots in coronary artery imaging.

Statistical pattern-recognition techniques have been frequently applied to the problem of medical diagnosis. Sequential Bayesian approaches are appealing because of the possibility of generating the underlying sensitivities, specificities, and prevalence statistics from the estimates of medical experts. The accuracy of these estimates and the consequences of inaccuracies carry implications for the future development of this type of system. In an effort to explore these subjects, the authors used statistics derived from a clinical database to revise the diagnostic logic in a Bayesian system for generating a differential diagnostic list. Substantial changes in estimated a priori probabilities, sensitivities, and specificities were made to correct for significant under- and overestimations of these values by a group of medical experts. The system based on the derived values appears to perform better than the original system. It is concluded that the statistics used in a Bayesian diagnostic system should be derived from a database representative of the patient population for which the system is designed.

To investigate whether a fixed external reference system should be used to most accurately describe regional left ventricular wall motion, we used vectors to analyze a simple model of ventricular contraction. If measured in a fixed external reference system, motion of implanted radiopaque midwall markers may contain translational as well as contractile components. Therefore, comparisons of different reference systems that use marker motion measured in a fixed reference system as a standard will be biased unless proper corrections are included. We conclude that evidence to data does not indicate that the use of a fixed reference system is superior to other methods for analyzing regional ventricular wall motion.

A prospective study of chest radiographic examinations in a respiratory intensive care unit was conducted to determine the diagnostic and therapeutic efficacy of such examinations. Analysis of data from 1,354 x-ray films from 167 patients revealed a 34.5 percent incidence of new (or increased) abnormalities, or tube or catheter malposition. Changes in diagnostic approach or therapeutic measures, excluding catheter position adjustments, occurred after 28.5 percent of the examinations. Radiographic yield was higher when a change in clinical condition prompted the radiographic examination than when the examination was a routine morning study. Changes in the approach to patient management were also more likely (42.7 percent) following examinations that were prompted by a change in a patient's clinical status. Less than 6 percent of the radiographic films taken post-procedure demonstrated abnormalities potentially related to the procedure. We conclude that, in a respiratory intensive care unit: routine morning radiographic examination frequently demonstrates unexpected or changing abnormalities, many of which prompt changes in diagnosis or management radiographic evaluation of a change in a patient's clinical condition has a higher yield than routine examinations; and post-procedure radiographic examination uncommonly demonstrates complications related to the procedure, but frequently demonstrates abnormalities of tube or catheter placement.

In domains where the types of data which are to be interpreted are relatively constrained (as in the case of specific laboratory test results), our modular data-driven approach can be very productive and well received by the clinical recipient of the data. The computer rarely surpasses the knowledge of an experts result from lack of communication, imperfect memory, oversight or multiple decision-makers caring for the same patient. In such cases, most of the alerts are immediately recognized as valid, so the need for elaborate explanations is not a high priority. On the other hand, a non-specialist is alerted to the need for additional investigation, tests or collaborative support, by the fact that a reminder or diagnosis that s/he had not previously considered, appears. In other words, for the expert, a data-driven system provides unceasing oversight in high-volume low-yielded situations where a small number of mistakes may uncommonly occur for reasons which are not related to the lack of knowledge of the provider. For the non-specialist the system suggests that the patient may have problems in a domain for which the physician needs additional support. In the present state of the art, we do not think that total reliance on the computer-contained knowledge is the ultimate source of this additional support; providing the awareness of the need may be the most important contribution. Once you know that you need help, it is usually obtainable. In a discussion about how computer systems have failed, Friedman and Gustafson made the following observation.(ABSTRACT TRUNCATED AT 250 WORDS)

Total lung capacity (TLC), functional residual capacity, residual volume, and corresponding 95% confidence intervals were measured in 245 healthy nonsmoking person (122 women, 123 men) using a single-breath helium technique. Prediction equations for lung volumes were generated by multiple linear regression. The resultant equations are similar to previously published equations using multiple-breath gas equilibration techniques. Measured 95% confidence intervals can be closely approximated by using two times the standard error of the estimate for each equation, but cannot be approximated by using +/- 20% of the predicted value. Radiographic TLC was not significantly different from the helium dilution TLC.

To assess the potential capability of using artery-specific factors to predict the success of coronary bypass grafting, we classified each graft according to the type of graft (mammary, single vein, or multiple skip vein), the degree of obstruction in the native coronary artery, and the size of the grafted artery. The mean patency rates for each category of graft were determined by performing early postoperative (mean 6.6 months) arteriograms in 354 patients. We found that mammary arteries gave the best results for every graft category. Skip vein grafts had a higher mean patency rate than single vein grafts for arteries with obstructions between 70% and 90% and lumina less than 2.0 mm in diameter. The results for vein and skip grafts were equivalent for the category with arteries greater than 2.0 mm and obstruction greater than 90%. Vein grafts produced better results than skip grafts for the remaining categories. Patients were classified into subgroups according to the number of actually observed graft failures. The probabilities that specific numbers of grafts would fail were calculated for each patient. These probabilities were based upon the observed patency rates for the category appropriate for each graft and the hypothesis that individual grafts within a patient fail independently. We observed an excess number of patients with no failures or multiple failures than would be expected according to the hypothesis of independent failure and category-specific patency rates. Conversely, there were fewer patients than expected who had single failures. Therefore, we postulate that there may also be global factors which influence early graft patency.

The safety of coronary bypass operations after coronary reperfusion with streptokinase for acute myocardial infarction is not well documented. Therefore we studied 23 consecutive patients (mean age, 59.5 years; 22 men) undergoing bypass operations a median of 5 days (range, 1 to 23 days) after thrombolysis (streptokinase). The control group consisted of 169 concurrent patients of similar mean age (58.8 years) having bypass operations for standard indications. The preoperative angiographic ejection fraction was 68 +/- 14% in the control patients and 61 +/- 14% in the streptokinase group (p less than 0.05). The number of diseased vessels (70% stenosis or greater) averaged 2.6 in control and 2.3 in streptokinase patients. A previous myocardial infarction had occurred in 42% of the controls and all of the streptokinase patients. Aortic cross-clamp times did not differ between the two groups (80 +/- 35 minutes for the controls and 68 +/- 25 minutes for the streptokinase group). Cardiopulmonary bypass times were similar: 108 +/- 45 minutes in the controls versus 109 +/- 28 minutes in the streptokinase group. Grafts per patient averaged 3.7 +/- 1.5 for the controls versus 2.8 +/- 1.1 for the streptokinase patients (p less than 0.01). Difficult operative hemostasis was noted in 4% of both groups. Inotropic support was given postoperatively to 11% of the control and 13% of the streptokinase patients (p = not significant). Measured blood loss during the first 48 hours postoperatively was similar, averaging 809 ml in controls and 776 ml in the streptokinase group. Blood product replacement was also comparable: mean, 713 ml in the control group versus 759 ml in the streptokinase group.(ABSTRACT TRUNCATED AT 250 WORDS)