Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

because observed behavior is the result of interactions between environmental factors and genes during the extended period of development. Therefore, to better understand expert and exceptional performance, we must require that the account specify the different environmental factors that could selectively promote and facilitate the achievement of such performance. In addition, recent research on expert performance and expertise (Chi, Glaser, & Farr, 1988; Ericsson & Smith, 1991a) has shown that important characteristics of experts' superior performance are acquired through experience and that the effect of practice on performance is larger than earlier believed possible. For this reason, an account of exceptional performance must specify the environmental circumstances, such as the duration and structure of activities, and necessary minimal biological attributes that lead to the acquisition of such characteristics and a corresponding level of performance. An account that explains how a majority of individuals can attain a given level of expert performance might seem inherently unable to explain the exceptional performance of only a small number of individuals. However, if such an empirical account could be empirically supported, then the extreme characteristics of experts could be viewed as having been acquired through learning and adaptation, and studies of expert performance could provide unique insights into the possibilities and limits of change in cognitive capacities and bodily functions. In this article we propose a theoretical framework that explains expert performance in terms of acquired characteristics resulting from extended deliberate practice and that limits the role of innate (inherited) characteristics to general levels of activity and emotionality. We provide empirical support from two new studies and from already published evidence on expert performance in many different domains.

/pdf/the-role-of-deliberate-practice-in-the-acquisition-of-expert-3j9768t5su.pdf

The role of deliberate practice in the acquisition of expert performance.

Part I. Experimental Studies: 2. Experiment in psychology 3. Experiments on perceiving III Experiments on imaging 4-8. Experiments on remembering: (a) The method of description (b) The method of repeated reproduction (c) The method of picture writing (d) The method of serial reproduction (e) The method of serial reproduction picture material 9. Perceiving, recognizing, remembering 10. A theory of remembering 11. Images and their functions 12. Meaning Part II. Remembering as a Study in Social Psychology: 13. Social psychology 14. Social psychology and the matter of recall 15. Social psychology and the manner of recall 16. Conventionalism 17. The notion of a collective unconscious 18. The basis of social recall 19. A summary and some conclusions.

Remembering. A Study in Experimental and Social Psychology, Cambridge (University Press) 1964.

Background
Decision aids are intended to help people participate in decisions that involve weighing the benefits and harms of treatment options often with scientific uncertainty.

Objectives
To assess the effects of decision aids for people facing treatment or screening decisions.

Search methods
For this update, we searched from 2009 to June 2012 in MEDLINE; CENTRAL; EMBASE; PsycINFO; and grey literature. Cumulatively, we have searched each database since its start date including CINAHL (to September 2008).

Selection criteria
We included published randomized controlled trials of decision aids, which are interventions designed to support patients' decision making by making explicit the decision, providing information about treatment or screening options and their associated outcomes, compared to usual care and/or alternative interventions. We excluded studies of participants making hypothetical decisions.

Data collection and analysis
Two review authors independently screened citations for inclusion, extracted data, and assessed risk of bias. The primary outcomes, based on the International Patient Decision Aid Standards (IPDAS), were:

A) 'choice made' attributes;

B) 'decision-making process' attributes.

Secondary outcomes were behavioral, health, and health-system effects. We pooled results using mean differences (MD) and relative risks (RR), applying a random-effects model.

Main results
This update includes 33 new studies for a total of 115 studies involving 34,444 participants. For risk of bias, selective outcome reporting and blinding of participants and personnel were mostly rated as unclear due to inadequate reporting. Based on 7 items, 8 of 115 studies had high risk of bias for 1 or 2 items each.

Of 115 included studies, 88 (76.5%) used at least one of the IPDAS effectiveness criteria: A) 'choice made' attributes criteria: knowledge scores (76 studies); accurate risk perceptions (25 studies); and informed value-based choice (20 studies); and B) 'decision-making process' attributes criteria: feeling informed (34 studies) and feeling clear about values (29 studies).

A) Criteria involving 'choice made' attributes:

Compared to usual care, decision aids increased knowledge (MD 13.34 out of 100; 95% confidence interval (CI) 11.17 to 15.51; n = 42). When more detailed decision aids were compared to simple decision aids, the relative improvement in knowledge was significant (MD 5.52 out of 100; 95% CI 3.90 to 7.15; n = 19). Exposure to a decision aid with expressed probabilities resulted in a higher proportion of people with accurate risk perceptions (RR 1.82; 95% CI 1.52 to 2.16; n = 19). Exposure to a decision aid with explicit values clarification resulted in a higher proportion of patients choosing an option congruent with their values (RR 1.51; 95% CI 1.17 to 1.96; n = 13).

B) Criteria involving 'decision-making process' attributes:

Decision aids compared to usual care interventions resulted in:

a) lower decisional conflict related to feeling uninformed (MD -7.26 of 100; 95% CI -9.73 to -4.78; n = 22) and feeling unclear about personal values (MD -6.09; 95% CI -8.50 to -3.67; n = 18);

b) reduced proportions of people who were passive in decision making (RR 0.66; 95% CI 0.53 to 0.81; n = 14); and

c) reduced proportions of people who remained undecided post-intervention (RR 0.59; 95% CI 0.47 to 0.72; n = 18).

Decision aids appeared to have a positive effect on patient-practitioner communication in all nine studies that measured this outcome. For satisfaction with the decision (n = 20), decision-making process (n = 17), and/or preparation for decision making (n = 3), those exposed to a decision aid were either more satisfied, or there was no difference between the decision aid versus comparison interventions. No studies evaluated decision-making process attributes for helping patients to recognize that a decision needs to be made, or understanding that values affect the choice.

C) Secondary outcomes

Exposure to decision aids compared to usual care reduced the number of people of choosing major elective invasive surgery in favour of more conservative options (RR 0.79; 95% CI 0.68 to 0.93; n = 15). Exposure to decision aids compared to usual care reduced the number of people choosing to have prostate-specific antigen screening (RR 0.87; 95% CI 0.77 to 0.98; n = 9). When detailed compared to simple decision aids were used, fewer people chose menopausal hormone therapy (RR 0.73; 95% CI 0.55 to 0.98; n = 3). For other decisions, the effect on choices was variable.

The effect of decision aids on length of consultation varied from 8 minutes shorter to 23 minutes longer (median 2.55 minutes longer) with 2 studies indicating statistically-significantly longer, 1 study shorter, and 6 studies reporting no difference in consultation length. Groups of patients receiving decision aids do not appear to differ from comparison groups in terms of anxiety (n = 30), general health outcomes (n = 11), and condition-specific health outcomes (n = 11). The effects of decision aids on other outcomes (adherence to the decision, costs/resource use) were inconclusive.

Authors' conclusions
There is high-quality evidence that decision aids compared to usual care improve people's knowledge regarding options, and reduce their decisional conflict related to feeling uninformed and unclear about their personal values. There is moderate-quality evidence that decision aids compared to usual care stimulate people to take a more active role in decision making, and improve accurate risk perceptions when probabilities are included in decision aids, compared to not being included. There is low-quality evidence that decision aids improve congruence between the chosen option and the patient's values.

New for this updated review is further evidence indicating more informed, values-based choices, and improved patient-practitioner communication. There is a variable effect of decision aids on length of consultation. Consistent with findings from the previous review, decision aids have a variable effect on choices. They reduce the number of people choosing discretionary surgery and have no apparent adverse effects on health outcomes or satisfaction. The effects on adherence with the chosen option, cost-effectiveness, use with lower literacy populations, and level of detail needed in decision aids need further evaluation. Little is known about the degree of detail that decision aids need in order to have a positive effect on attributes of the choice made, or the decision-making process.

/pdf/decision-aids-for-people-facing-health-treatment-or-1j1706q17l.pdf

Decision aids for people facing health treatment or screening decisions

Background Audit and feedback continues to be widely used as a strategy to improve professional practice. It appears logical that healthcare professionals would be prompted to modify their practice if given feedback that their clinical practice was inconsistent with that of their peers or accepted guidelines. Yet, audit and feedback has not been found to be consistently effective. Objectives To assess the effects of audit and feedback on the practice of healthcare professionals and patient outcomes. Search strategy We searched the Cochrane Effective Practice and Organisation of Care Group's register up to January 2001. This was supplemented with searches of MEDLINE and reference lists, which did not yield additional relevant studies. Selection criteria Randomised trials of audit and feedback (defined as any summary of clinical performance over a specified period of time) that reported objectively measured professional practice in a healthcare setting or healthcare outcomes. Data collection and analysis Two reviewers independently extracted data and assessed study quality. Quantitative (meta-regression), visual and qualitative analyses were undertaken. Main results We included 85 studies, 48 of which have been added to the previous version of this review. There were 52 comparisons of dichotomous outcomes from 47 trials with over 3500 health professionals that compared audit and feedback to no intervention. The adjusted RDs of non-compliance with desired practice varied from 0.09 (a 9% absolute increase in non-compliance) to 0.71 (a 71% decrease in non-compliance) (median = 0.07, inter-quartile range = 0.02 to 0.11). The one factor that appeared to predict the effectiveness of audit and feedback across studies was baseline non-compliance with recommended practice. Reviewer's conclusions Audit and feedback can be effective in improving professional practice. When it is effective, the effects are generally small to moderate. The absolute effects of audit and feedback are more likely to be larger when baseline adherence to recommended practice is low.

Audit and feedback: effects on professional practice and healthcare outcomes

We examine three related ways in which the gap between theory and practice has been framed. One approach views it as a knowledge transfer problem, a second argues that theory and practice represent distinct kinds of knowledge, and a third incorporates a strategy of arbitrage--leading to the view that the gap is a knowledge production problem. We propose a method of engaged scholarship for addressing the knowledge production problem, arguing that engaged scholarship not only enhances the relevance of research for practice but also contributes significantly to advancing research knowledge in a given domain.

Knowledge for Theory and Practice

Providing explanations for recommended actions is deemed one of the most important capabilities of expert systems (ES). There is little empirical evidence, however, that explanation facilities indeed influence user confidence in, and acceptance of, ES-based decisions and recommendations. This paper investigates the impact of ES explanations on changes in user beliefs toward ES-generated conclusions. Grounded on a theoretical model of argument, three alternative types of ES explanations A trace, justification, and strategy A were provided in a simulated diagnostic expert system performing auditing tasks. Twenty practicing auditors evaluated the outputs of the system in a laboratory setting. The results indicate that explanation facilities can make ES-generated advice more acceptable to users and that justification is the most effective type of explanation to bring about changes in user attitudes toward the system. These findings are expected to be generalizable to application domains that exhibit similar characteristics to those of auditing: domains in which decision making tends to be judgmental and yet highly consequential, and the correctness or validity of such decisions cannot be readily verified.

https://www.jstor.org/stable/pdfplus/249686.pdf

The impact of explanation facilities on user acceptance of expert systems advice

A structured review of the effect of economic incentives on consumers' preventive behavior.

PURPOSE We wanted to assess the impact of an electronic health record–based diabetes clinical decision support system on control of hemoglobin A1c (glycated hemoglobin), blood pressure, and low-density lipoprotein (LDL) cholesterol levels in adults with diabetes.

METHODS We conducted a clinic-randomized trial conducted from October 2006 to May 2007 in Minnesota. Included were 11 clinics with 41 consenting primary care physicians and the physicians’ 2,556 patients with diabetes. Patients were randomized either to receive or not to receive an electronic health record (EHR)–based clinical decision support system designed to improve care for those patients whose hemoglobin A1c, blood pressure, or LDL cholesterol levels were higher than goal at any office visit. Analysis used general and generalized linear mixed models with repeated time measurements to accommodate the nested data structure.

RESULTS The intervention group physicians used the EHR-based decision support system at 62.6% of all office visits made by adults with diabetes. The intervention group diabetes patients had significantly better hemoglobin A1c (intervention effect −0.26%; 95% confidence interval, −0.06% to −0.47%; P=.01), and better maintenance of systolic blood pressure control (80.2% vs 75.1%, P=.03) and borderline better maintenance of diastolic blood pressure control (85.6% vs 81.7%, P =.07), but not improved low-density lipoprotein cholesterol levels (P = .62) than patients of physicians randomized to the control arm of the study. Among intervention group physicians, 94% were satisfied or very satisfied with the intervention, and moderate use of the support system persisted for more than 1 year after feedback and incentives to encourage its use were discontinued.

CONCLUSIONS EHR-based diabetes clinical decision support significantly improved glucose control and some aspects of blood pressure control in adults with type 2 diabetes.

https://www.annfammed.org/content/annalsfm/9/1/12.full.pdf

Impact of electronic health record clinical decision support on diabetes care: A randomized trial

https://onlinelibrary.wiley.com/doi/pdf/10.1207/s15516709cog0503_3

Paul E. Johnson

Papers

Knowledge for Theory and Practice

The impact of explanation facilities on user acceptance of expert systems advice

A structured review of the effect of economic incentives on consumers' preventive behavior.

Impact of electronic health record clinical decision support on diabetes care: A randomized trial

Expertise and error in diagnostic reasoning