Development and significance of automated history‐taking software for clinical medicine, clinical research and basic medical science

The best we can do is to divide all processes into those things which can be better done by machines and those which can be better done by humans. The meaning of von Neumann's statement is that rule-based tasks are undertaken more faithfully by computers than by humans and that the potential benefit of computing can be assessed by evaluating whether there is less than acceptable adherence to applicable rules when a task is carried out by humans. von Neumann's prescient idea has been applied to essentially all theoretical, experimental and applied sciences and engineering. Today, computing is indispensable to these disciplines and fuels their progress. However, there has been little application of computing to the fundamental, rule-based processes underlying effective practice of medicine, specifically to collecting and interpreting clinical data from patients. The nature of information collected from examining patients is different, of course, from data in physical experiments. However, the processes for collecting these data, and their interpretation, are no less rule based. The absence of significant computing power in daily clinical practice cannot be attributed to lack of need as, whenever scrutinized, the processes by which physicians collect data from patients and make clinical decisions are noncompliant with applicable rules 1-10. These failures contribute to less than acceptable outcomes for patients 11-14 and to as many as 98 000 deaths/year in the USA from medical errors 11. Poor data collection in clinical practice also impedes progress in clinical 15-18 and basic science research 19, 20. The purpose of this review is to summarize the development of automated history-taking software, considering the evidence for the validity and superiority of this method for collecting clinical data, and the continued need for history taking as the basis for quality care. Clinicians unfamiliar with programs for automated history taking need to begin to understand the power of the technology and its potential for improving clinical outcomes. Physicians are familiar with electronic health record (EHR) technology; however, EHRs do not incorporate formalized rules for history taking, cannot improve adherence to these rules and hence lack the benefits that come from automating rule-based tasks. True automation of history taking may be considered ‘the second generation of EHRs’ and, in our opinion, will be an inevitable development. This is the only mechanism for removing the incompatibility between unaided, finite human cognition and limited physician time and maximizing the value of healthcare at affordable cost. The knowledge base for medical practice is enormous, and is expanding continuously 21 and growing in complexity. Furthermore, established facts are subject to change. Apparently simple complaints can have extensive differential diagnoses. These cognitive loads are exacerbated by less time to see more patients. The profession's organized response to the cognitive challenges has been specialization so that each specialist manages repetitively a relatively narrow set of clinical issues. Specialization, however, has not improved clinical outcomes for patients, which are superior in geographical areas with higher proportions of primary care physicians 22-25. An alternative to specialization is the use of computing to reduce the cognitive load of practice. Efforts in this direction began in the late 1940s. Brodman et al. 26 developed the Cornell Medical Index as a questionnaire to standardize data collection, orientated towards systems review to provide ‘… a quick and reliable method of obtaining important facts about a patient's medical history without expenditure of the physician's time’. It may seem trivial, but the Index demonstrated that history taking did not require a physician. In fact, the Index collected significantly more relevant information than physicians examining the same patients. Subsequently it was shown that a computer, when fed standardized, self-reported history data, was as accurate as physicians in diagnosing a wide array of common disorders 27. This work thus dispelled the then prevalent notion that diagnosis was ‘… indefinable and intuitive…’ rather than ‘… logical and completely defined…’ 27. The value of self-reported medical histories collected by questionnaire and processed by computer was replicated at Kaiser Permanente clinics using a questionnaire of 600 questions 28. Results were reported only for the diagnosis of asthma. Sensitivity and specificity of the analytical method were good. Slack and colleagues expanded the reach of self-reported history taking by programming a computer to collect the history directly from patients 29. The rules for history taking were programmed by branching question trees so that the path from one question to another was determined by a prior answer and could be related to different, specific medical issues in different patients. This simple logic function made it possible to expand the number of questions posed by computer as compared with a paper questionnaire. This first computerized history-taking program was limited to allergy-related issues but was extended to separate programs for the review of systems and several problem-specific interviews 30. The success of these programs was measured by the completeness of data collection as compared with physician-acquired histories, with which they compared favourably. Mayne and colleagues at the Mayo Clinic were also early developers of computerized history-taking programs. Their program asked patients to select a chief complaint from a menu of complaints 31. The program contained no follow-up to resolve the differential diagnosis of the chief complaint but encompassed a review of systems plus elements of the past medical history 31. A program with clinical coverage similar to that developed at the Mayo Clinic was tested in a small number of in-patients by Grossman et al. 32 with similar results, i.e. far more clinical information was recorded by computers than by physicians. Lawrence Weed was another pioneer in the field. Weed's starting premise was different from that of his predecessors. He explicitly acknowledged that the knowledge for practice exceeded what anyone could learn and use for collecting relevant clinical data from patients and for interpreting the data to make accurate, timely diagnoses and that computing was a method for bridging this gap 33, 34. His purpose was to improve decision-making through better data collection coupled to automated decision support. Weed's history taking was committed to diagnosis by the closeness of fit between the set of clinical attributes displayed by a given patient with a specific chief complaint and the sets of attributes displayed by patients covering all the entities in the differential diagnosis of that chief complaint. For example, for a patient with vertigo/dizziness as the chief complaint, Weed's scheme queried every patient for the presence of all attributes that could be displayed by any patient with any of the 78 diagnostic entities in his differential diagnosis for vertigo/dizziness 35. The key output from the Weed program was not the facts of the history but the closeness of fit between the attributes of an interviewed patient and the attributes of a patient with each of the entities in the differential diagnosis of the patient's chief complaint. Weed termed the history-taking routines ‘Problem Knowledge Couplers’. The ‘problem’ was the patient's chief complaint, and the ‘knowledge’ was the database of attributes of well-characterized patients with known diagnoses. Published work with couplers does not cite a source for the ‘Knowledge’ databases. Furthermore, the overall scheme was not tested rigorously for utility and validity. A single clinical trial of the coupler program failed to establish diagnostic value as compared with routine care 36. The patient population tested was relatively young, however, so this negative result does not mean that coupler programs are without value. Couplers have been reportedly used since 2005 only in a single small practice, but without data to assess how they are used and how they contribute to management 37. The coupler programs are now owned by AskMD 38. It seems unlikely that the couplers are being used as originally intended. Many other investigators developed computerized history-taking programs during this period until 2000. A summary of these programs is provided in table 3 of the review by Bachman 39. However, none of the programs cited expanded the technology beyond the range of issues addressed by Slack et al., Mayne et al. and Weed. Self-reported history entries in the Cornell Medical Index questionnaire were documented by re-interview of patients and found to be highly accurate 26. Discrepancies between self-reported entries for computer interview and physician records reflected equally low error rates by computer and physician 40. Notably, however, computerized histories included more useful clinical data as compared with physician histories 40. Analysis of data collected by self-reported, computerized history taking for pre-operative clearance showed that 97% of retest responses from 239 patients were identical to test answers 41. A similar study of 48 patients interviewed using a general medical history program, showed a test–retest replication rate of 93% 42. A quarter of the small number of discrepancies were patient errors in selecting the intended answers; another quarter occurred because patients were uncertain which of the available answers applied to them. Because self-reported clinical data acquired by computerized history taking are accurate, there is more clinically significant information in these histories as compared with medical charts. Indeed, comparisons of the accuracy of self-reported data and medical records 1, 2, 5, 7-9 show that the former are superior. This is not unexpected because computers do not interpret self-reported answers. There is no control over this aspect of history taking by physicians. Moreover, use of surrogates and scribes 43 in the era of EHRs makes entry of accurate clinical data into the record increasingly complex in ways avoided by automating the process. Even at a time when few had experience with or knowledge of computers, patients had a high level of acceptance of the technology for history taking 29, 31, 44, 45. This remains true now that computing is ubiquitous 42, 46. Patient acceptance of computerized history taking extends beyond willingness to use these programs to belief that use will improve the quality of their healthcare 32. Patients are also quick to humanize the computer 47, 48. In addition, patients are willing to invest amounts of time inputting data that are not available to physicians, e.g. self-reported history-taking sessions exceeding 1 h 31, 42. Another advantage of computerized history taking is that patients may reveal information about risky lifestyles to the computer that they will not share with physicians 49, 50. It has not been tested for users of computerized history-taking programs; but studies of the impact of paper questionnaires on physicians show that patients using questionnaires are more satisfied with their care than those not providing the physician with self-reported clinical information 51. There has been no significant physician ‘buy-in’ to the use of computerized history-taking programs 31, 41 Physicians who were questioned about the value of these programs believed that differences between their and computer findings were false-positive computer outputs or nonrelevant information 31, 41. Quaak et al. 52 reported a similar view. Physicians who were not responsible for primary data collection were asked to compare the diagnostic value of information acquired by a computerized, general medicine history-taking program with information from physician-acquired histories 52, 53. Although the computerized histories were acknowledged to contain more clinical information, physician histories were deemed more valuable for diagnoses. the test–retest of automated histories, physician of high rates of false-positive findings in self-reported histories are not It seems that physicians the value of collecting more clinical information than they This is to the negative impact of limited human cognition on clinical the information a physician can record in a as an and use to make a management decision is limited at all by a finite cognitive about the of data to a clinical decision on their cognitive with an to by and time A computer has a for and of of not subject to by and limited computerized history taking more clinically relevant information than a physician can process. The physician will of this as to decision-making when the information is to the or unfamiliar of This between what computers can output and what physicians can make of is a common for example, are ‘… by their The value of automated history taking cannot be by physician about the data collected by computer but on the impact of the information on outcomes for patients. history taking collect and output information that a physician will as and findings to their to a specific differential diagnosis and in this as they during an For example, when an entry a programmed the program can a question to for false-positive and Another method for improving automated history taking is to reduce the cognitive load of to physicians by findings as This method of complex information compared with a of to the of complex information in the negative cited by Quaak et al. It is to the meaning of data in the output from their program of the programs cited by Bachman may be in use at in which they were but none seems to be in use for Medical programs developed more by the with patients on single issues and seem to have had limited use their of development. An is self-reported of by patients with in In addition, The of is a clinically automated history-taking application to a ‘… the and and ‘… processes the The of to to clinical outcomes by using analytical to and to the power of data to improve its There is no program the medical This history-taking program how the of practice may be by using computer technology to limited Slack et al. have expanded on work to a general medicine, history-taking However, no about its have been is it that the program is in Medical is as a method by which the patient's answers data in the physician's of coverage and are not The program may be the only automated history-taking program development for medical histories from other history-taking in the in that it clinical continuously as data are The key to this is automated of all prior answers as the interview so that the questions are determined by that all prior answers to the differential This mechanism also for entry errors and answers. 1 is an of the decision that the of in a the logic to a through the is in the is a of the more than decision of the program at patient by selecting a chief complaint at the by the and the of the of about 000 decision The logic which questions are relevant to and thus posed to a specific and standardize data collection all issues and for all interviews by that questions not asked are nonrelevant to the patient's medical issues. There is no to the number of that can be in and the range of clinical and issues the program can to the to findings as differential management and for diagnostic and and This is another important between and other automated which only physicians using the of significant clinical 52, which is a of differences between computer and physician in clinically significant also physicians in adherence to which was measured for the management of data elements as This and of data sets that are true for the data in a given interview are also as These can be by computer to improve for complex clinical Furthermore, because clinical management is a automated history-taking programs have to function time to provide automated data collection relevant to This on to in the of and clinical attributes to in the towards or of automated follow-up in the of and can incorporate physical and for this is owned by a for which the is by the Knowledge and to from the use of this knowledge are by the medical history taking the time and cognitive for collecting clinical data from the physician to the patient and a The technology can data may be for automated decision a wide range of clinical issues. a single of the to incorporate knowledge to of clinical to practice. by using a the collection of clinical data can be all patients in with the healthcare and included in a database to clinical and basic medical to it is possible to the of medical knowledge on a that can be for a few There are with which anyone who can use a program can medical knowledge as with are in affordable care of development and automated history taking is essentially not used in medical practice. evidence that physicians who developed these programs more on what was possible rather than on programs that be by the clinical use on physicians to the for automated history taking, and and of the medical knowledge into the software, to the value of computing power and a of clinical The clinical and patient be in these In addition, for clinical establish that automated history taking can impact the only that clinical outcomes for patients. our only one trial has been to that measured an on In this established patients in primary care with Medical from Physicians determined from these data, available that of patients did not need care. Their medical issues and how they from patients deemed to need care were not However, this is the only clinical reported so far to be by the use of automated history automated history taking to clinical use remains a an of specialist and primary care physicians, patient computer and experience A significant has been made in this by the development of the development and of as a the medical and are of similar at any other physicians will to to the He is the Medical this and of the of a and its management begin with the to a In to these are considered less important than in the past and when are directly of for patients Physicians are data for history taking In addition, it is believed that medical practice is about to be by and related that will provide all the information to make diagnoses and the in taking a history of to the value of medical knowledge for patient there be benefit from providing information at in value to the of patients. tested in specific however, the superiority of as compared with examining the patient cannot be there is an between the use of and quality of outcomes for patients In addition, to be useful for and this application is to there is no experimental for a in medical practice. and provide information on the of function and result from and of between these between complex systems in and between and course, can be in however, it is in to and for the of in of A in a specific can have for different systems of which it is a The on function from a in any given can on the of all other of the i.e. is studies of or a few in cannot between of into of and the of the on these In other of and are to from data ‘… there is no to all the relevant data about each included in the In with level information, information from clinical on the overall of complex i.e. whether or as as the array of and in an and how these with The best is history taking the key data set for decision-making to the quality of care for patients and that this will not for the Indeed, the significant in practice since time is the of the knowledge that practice, for effective more information about the patient than in the taking is the only aspect of clinical data collection not from the of how to value. of history taking in practice makes it possible to study this by what information to be collected in which patients. the of to the will This result that outcomes in are by or that of cannot be determined for with better than about can test whether this is a true on knowledge for as a routine of practice by automated history 1 and show to that are not in and that are not included in The are for complex For example, for information, by on a set of complex processes that cannot be in In addition, basic that the complex systems on which the are unlikely to function of each There are of at of many taking history taking can data shown in 1 and without physician time and the data for without physicians with clinical facts to decision-making at the ‘… to and to summarize the data in ways can be applied to the complex data sets that will be for can be to determined by data these improve on an data sets to this can be and with the data elements a clinical decision to can be to physicians on a This general scheme is applicable to any for which there are no or clinical research can an of routine care history taking can every practice and healthcare to the of clinical history-taking also will have a on the progress of basic medical research 19, 20. The key to the quality of clinical data for with findings at the level is the of the data and its of and and by potential of to of and history of and history past with of or history and with at of of daily of to to other than at at diagnosis to to other than There is a that EHRs will have an impact on clinical and basic science research similar to that for automated history taking data cannot a of clinical Because of the time of use compared with or data into EHRs may be collected by physicians, by with or by scribes Physicians using EHRs do not collect data by a or with a Furthermore, the of data in EHRs is for many other The physician the output from an automated history-taking program is in the of a providing a second This will with patients by removing the of data collection and providing time to in more the patient's issues and In addition, automated history taking will the use of which can with patient Physicians have medicine, in general and for in these computer technology does not provide patients with into the processes of practice. computers provide patients with a of a key by which medicine is that they to use of automated history-taking in prior work may be in this the power of automated history-taking is for better outcomes at affordable this question to be by in and development of the technology as it It will also be important to that physicians, have finite cognitive and that medical practice is an cognitive it may be important to the between and computerized history provide a better of the and than the The history-taking a more of the of than is available to the are only to a physician. is in the is evidence for better clinical outcomes from the use of history-taking This is for of automated history-taking programs. from history taking and of patients remains for quality outcomes. with in and related will not data collection in practice. Indeed, the range of information available from interview of patients that be into clinical decisions is Furthermore, in to the cognitive of history taking, there is less time for data These in routine practice can be addressed with computer technology because history taking is a rule-based It is established that computers can medical histories by interview of patients, that automated history taking more clinically significant information than physicians examining the same patients and that the data are have a response to interview by is to automated data collection to any of to clinical and to computers to be of and with the self-reported history taking is a key technology for maximizing outcomes for patients, and will improve the It is also increasingly that clinical practice and clinical research be to improve the knowledge base for practice, which can only be through automated history no of is the on for technology related to the program that are without any to in related to the development of have been in by the The for programming in and colleagues at the for in the program, with to and The is to for his of the program at and for his and of the