An Introduction to Machine Learning

In the last few years, machine learning (ML) and artificial intelligence have seen a new wave of publicity fueled by the huge and ever-increasing amount of data and computational power as well as the discovery of improved learning algorithms. However, the idea of a computer learning some abstract concept from data and applying them to yet unseen situations is not new and has been around at least since the 1950s. Many of these basic principles are very familiar to the pharmacometrics and clinical pharmacology community. In this paper, we want to introduce the foundational ideas of ML to this community such that readers obtain the essential tools they need to understand publications on the topic. Although we will not go into the very details and theoretical background, we aim to point readers to relevant literature and put applications of ML in molecular biology as well as the fields of pharmacometrics and clinical pharmacology into perspective.

Clinical pharmacology = disease progression + drug action

Clinical pharmacology is concerned with understanding how to use medicines to treat disease. Pharmacokinetics and pharmacodynamics have provided powerful methodologies for describing the time course of concentration and effect in individuals and in populations. This population approach may also be applied to describing the progression of disease and the action of drugs to change disease progress. Quantitative models for symptomatic and disease-modifying effects of drugs are valuable not only for describing drugs and diseases but also for identifying criteria to distinguish between types of drug actions, with implications for regulatory decisions and long-term patient care.

Drug-target interaction prediction by integrating chemical, genomic, functional and pharmacological data

In silico prediction of unknown drug-target interactions (DTIs) has become a popular tool for drug repositioning and drug development. A key challenge in DTI prediction lies in integrating multiple types of data for accurate DTI prediction. Although recent studies have demonstrated that genomic, chemical and pharmacological data can provide reliable information for DTI prediction, it remains unclear whether functional information on proteins can also contribute to this task. Little work has been developed to combine such information with other data to identify new interactions between drugs and targets. In this paper, we introduce functional data into DTI prediction and construct biological space for targets using the functional similarity measure. We present a probabilistic graphical model, called conditional random field (CRF), to systematically integrate genomic, chemical, functional and pharmacological data plus the topology of DTI networks into a unified framework to predict missing DTIs. Tests on two benchmark datasets show that our method can achieve excellent prediction performance with the area under the precision-recall curve (AUPR) up to 94.9. These results demonstrate that our CRF model can successfully exploit heterogeneous data to capture the latent correlations of DTIs, and thus will be practically useful for drug repositioning. Supplementary Material is available at http://iiis.tsinghua.edu.cn/~compbio/papers/psb2014/psb2014_sm.pdf.

What do clinical pharmacologists do? A questionnaire survey of senior UK clinical pharmacologists

Since the discipline expanded during 1970-1990, the number of UK consultant clinical pharmacologists has fallen. This paper describes the results of a questionnaire survey of the work patterns of 53 UK consultant clinical pharmacologists, including 35 (66%) employed by universities and 13 (25%) employed by the National Health Service (NHS). The range of activities undertaken includes: teaching medical students and others the principles of clinical pharmacology and practical therapeutics; research in a wide range of pharmacological and therapeutic areas; patient care, mostly in acute and general hospital medicine and in out-patient clinics; service both locally and nationally on a wide range of committees related to drug therapy; editorial work on learned journals and preparation of written teaching materials, including journal articles, didactic textbooks, reference books, and e-learning materials. The median amount of time that a UK consultant clinical pharmacologist spends on these activities is 50 hours per week; several work more than that. The time is spent as follows: teaching 10%; research 40%; clinical work 30%; policy and administration 12%; editorial work and writing 8%. The numbers of physicians who have been newly registered with the General Medical Council as clinical pharmacologists each year has so far been undiminished, but the number of consultant posts has fallen in 1990-2010. Many trainees therefore leave the discipline when they become consultants.

Computing normalised prediction distribution errors to evaluate nonlinear mixed-effect models: the npde add-on package for R

Pharmacokinetic/pharmacodynamic data are often analysed using nonlinear mixed-effect models, and model evaluation should be an important part of the analysis. Recently, normalised prediction distribution errors (npde) have been proposed as a model evaluation tool. In this paper, we describe an add-on package for the open source statistical package R, designed to compute npde. npde take into account the full predictive distribution of each individual observation and handle multiple observations within subjects. Under the null hypothesis that the model under scrutiny describes the validation dataset, npde should follow the standard normal distribution. Simulations need to be performed before hand, using for example the software used for model estimation. We illustrate the use of the package with two simulated datasets, one under the true model and one with different parameter values, to show how npde can be used to evaluate models. Model estimation and data simulation were performed using NONMEM version 5.1.

Pharmacometrics and the transition to model-based development

As the transition to model-based drug development continues, pharmacometric analysis will have an increasingly important role across the entire life cycle of drug discovery, development, regulatory approval, and commercialization. For this reason, pharmacometrics can--and should--have an integrating function in the transformation to model-based development. This essay describes an approach for formalizing the pharmacometrics process using the disciplines encompassed by enterprise engineering.

Early discontinuation: more frequent among general practitioners with high levels of prescribing

INTRODUCTION

Discontinuation of medical drug treatment is a serious problem in primary care. The need for a better understanding of the processes, including physician-specific mechanisms, is apparent. The aim of this study was to analyse the association between general practitioners' prescribing prevalence and rate of early discontinuation of different drugs consisting of, in this study, lipid-lowering drugs, antihypertensive drugs, antidepressants, antidiabetics and drugs against osteoporosis.

MATERIAL AND METHODS

This was a register study based on prescription data covering a 4-year period and consisting of 470,000 citizens. For each practice and group of drug, a 1-year prevalence for 2002 and the rate of early discontinuation among new users in 2002-2003 were estimated. Early discontinuation was defined as no prescriptions during the second half-year following the first prescription, with the exception of new users of antidepressants for whom it was the first half-year. Correlations were analysed using the SAS ver. 9.1PROC MIXED procedure. The association with the total prescribing prevalence (all drugs) was also analysed.

RESULTS

A total of 141 general practices were included in the study. There was a positive association between the prevalence of prescribing for the specific drugs studied (antidepressants, antidiabetics, drugs against osteoporosis and lipid-lowering drugs) and early discontinuation (r = 0.29 -0.44), but not for anti-hypertensive drugs. The analysis of the association between prevalence of all drugs and drug-specific early discontinuation showed some degree of positive association - strongest for anti-hypertensive drugs (r = 0.62) and antidepressants (r = 0.43).

CONCLUSION

This study confirmed our hypothesis that general practitioners with high levels of prescribing attain higher rates of early discontinuation compared with colleagues with low levels of prescribing, not only with respect to antidepressants but also for various groups of drugs. A common underlying mechanism is suggested but has to be verified in future studies.

Potential drug interactions during a three-decade study period: a cross-sectional study of a prescription register

OBJECTIVES

The increased risk of adverse events in patients receiving potentially interacting drugs has long been recognized. The purpose of the present study was to evaluate the change in the risk of receiving potentially interacting drugs during a period covering three decades and to examine the relative risk of actual drug combinations.

METHODS

The prescriptions from all individuals (about 8,000) with two or more prescriptions during three periods of 15 months, October to December 1983-1984, 1993-1994 and 2003-2004, were collected from an ongoing cohort study in the county of Jämtland, Sweden. The potential interactions were detected by a computerized system.

RESULTS

The relative risk (RR) of receiving potentially interacting drugs increased for type C interactions [RR: 1.177, 95% confidence interval (CI): 1.104-1.256] and decreased for type D interactions (RR: 0.714, 95% CI: 0.587-0.868) from the period 1983-1984 to 2003-2004. Polypharmacy for the participants increased by 61%, from 9.05 filled prescriptions per subject in 1983-1984 to 10.6 in 1993-1994 and 14.6 in 2003-2004. The RR was positively correlated to the pronounced increase in polypharmacy; in addition, an exponential relationship was found for the more severe type D interactions. Few interacting drug combinations were responsible for a large proportion of the risk.

CONCLUSION

We conclude that the risk of receiving potentially interacting drugs was strongly correlated to the concomitant use of multiple drugs. The pronounced increase in polypharmacy over time implies a growing reason for prescribers and pharmacists to be aware of drug interactions. Recently established national prescription registers should be evaluated for drug interaction vigilance, both clinically and epidemiologically.

Description of controlled trials published in Methods and Findings, 1979-2002

It is thought that the controlled trial (CT) is the most adequate research method to assess a therapeutic intervention in terms of efficacy, and it also constitutes the basis for the development of systematic reviews on health interventions. To identify and obtain the majority of published CTs is not an easy task, mainly because of limitations concerning the currently available electronic sources. The aim of the present work was to identify, describe, and assess the quality of CTs published in the journal Methods and Findings in Experimental and Clinical Pharmacology (M&F). Additionally, to assess the retrievability of both methods, a search was performed in Medline (PubMed access) through the use of an optimal search strategy for CTs. A total of 189 original studies out of a total of 2796 reviewed articles met the CT criteria according to the Jadad scale score, we could hold that only 58% of the CTs were of good quality. The present work confirms, once again, the limitations of a CT search performed exclusively through Medline (sensitivity 64% and specificity 98%). In conclusion, we suggest that the journal M&F explicitly joins the International CONSORT Statement.

Building an ontology of adverse drug reactions for automated signal generation in pharmacovigilance

Automated signal generation in pharmacovigilance implements unsupervised statistical machine learning techniques in order to discover unknown adverse drug reactions (ADR) in spontaneous reporting systems. The impact of the terminology used for coding ADRs has not been addressed previously. The Medical Dictionary for Regulatory Activities (MedDRA) used worldwide in pharmacovigilance cases does not provide formal definitions of terms. We have built an ontology of ADRs to describe semantics of MedDRA terms. Ontological subsumption and approximate matching inferences allow a better grouping of medically related conditions. Signal generation performances are significantly improved but time consumption related to modelization remains very important.

It's time

Statistical inference involves taking the results of models and knowledge about probability to make decisions about the relationship in question. This commentary explains the usefulness of statistical inference to the drug development process, as well as some common pitfalls. It also examines reasons why statistical inference does not seem to be fully integrated into pharmacometric modeling. An example is shown that demonstrates the inferential advantages of mechanistic models. Both statisticians and pharmacometricians ought to take note of these advantages and integrate their efforts in order to maximize the decision-making potential of clinical research.

Pediatric resident education and needs assessment in clinical pharmacology

OBJECTIVE

To identify perceived and unperceived educational needs of residents to organize a seminar series in clinical pharmacology.

METHOD

All pediatric residents (48) and all attending general pediatric staff (20) were sent structured questionnaires with potential seminar topics. Data from previous pharmacy chart audits and complaints lodged with patient care representatives were analyzed as the environmental scans.

RESULTS

There was a 75% response rate from both residents and staff. The responses were very similar and the only significant difference was the response to a seminar on correcting electrolyte imbalances which the residents favoured (p = 0.005). The environmental scans identified pain management as one of the main areas needing improvement.

CONCLUSION

Perceived learning needs of residents are similar but not identical to those identified by the faculty. Environmental scanning can be used to identify unperceived learning needs.

Stimulus- and event-related evoked potentials: from neuroscience to clinical neuropharmacology

Synaptic neural (and neural system) functions are peculiarly sensitive to neuroactive compounds. Pharmacological interference/modulation is readily reflected by modifications in the organization of central nervous system (CNS), electrophysiologic signals occurring spontaneously in response to sensory stimulation (stimulus-related or evoked responses) or elicited in conjunction with sensory, motor or cognitive events (event-related potentials). Evoked responses reflect the basic physiology of sensory processes, while event-related potentials combine the time/space resolution of electrophysiologic signals with the specificity of eliciting neuropsychological conditions. The rationale for investigating drug effects on evoked and event-related potentials is manifold. Both are related to sensory and operant behavior and under suitable experimental conditions allow interpretation of drug-related changes in terms of CNS excitability. Some continuity between observations in man and in vivo or in vitro animal data is often possible. Proper handling of the stimulus physical properties or experimental/situational links may allow the responses to be related to sensory input or to neuropsychological manipulation of selectively activated CNS functions or functional subsystems and therefore to control spontaneous variability. This review summarizes today's knowledge of the application of electrophysiology to human neuropharmacology, with due reference to basic pharmacology and experimental evidence.

Subjects presented at ASHP Midyear Clinical Meetings, 1967-90

The subject matter and trends of presentations made at ASHP Midyear Clinical Meetings (MCMs) were studied. A computerized database of information from MCM and Annual Meeting (AM) program and abstract books was created. The data were analyzed to determine the distribution of MCM presentations by subject and by author for the period 1967 to 1990, determine if there were differences in subjects covered between MCMs and AMs, explore the proposition that there has been duplication of material in MCM presentations, and evaluate the frequency with which MCM presentations have been published in ASHP journals. The total number of presentations made at MCMs from 1967 through 1990 was 8180, while the total for the AMs was 1547 for the two periods (1962-71 and 1985-90) studied. The most common keywords in titles were "pharmacy," "drug," "patient," "hospital," and "service." All International Pharmaceutical Abstracts subject categories and two other categories were represented; presentations in the institutional pharmacy practice category were the most frequent, while pharmacognosy-related presentations were least frequent. The overwhelming number of authors made only one presentation and were listed as the first author. The subjects of presentations were similar between AMs and MCMs. A tendency toward duplication of material was found. Of the 8180 MCM presentations, at least 1005 were published in an ASHP journal. Between 1967 and 1990, presentations at MCMs covered a wide range of subjects but were sometimes duplicative or not on the cutting edge.

Survey of Canadian oncology pharmacy services--the emergence of a specialty

The provision of pharmacy services to oncology is a substantial component of hospital pharmacy practice at several Canadian hospitals. To determine the scope of such pharmacy services a survey was developed and distributed to 103 hospital pharmacies and 11 ambulatory oncology pharmacies in 1988. There were 72 responses (65%), although only 96 centres were known to offer oncology services (adjusted response = 72 of 96 = 75%). Surveys were distributed and returned in reasonable proportion across the country. Inpatient and outpatient services were reported separately, for size of service offered, and categories of work in which staff were employed. Twenty-four of fifty-seven centres report large inpatient pharmacist services, and 21 of 57 had large inpatient technician services. The ratio of pharmacists to technicians appeared to be about equal, but technicians were less likely than pharmacists to be permanently assigned (one-third vs. one-half permanently assigned). About one-half of both large and small services indicate a desire for increased time for both pharmacists and technicians. Outpatient services were reported by fewer respondents, and the job assignments in this setting were mostly permanent. Manpower usage in both settings is primarily dedicated to drug preparation and distribution, although two-thirds of centres report small clinical services (most centres desired increased clinical services). Future planning topics ranked improved clinical services and standardization of practice highest. A 91% majority agree that there should be standards for pharmacy practice in oncology pharmacists in Canada. Many factors, including insufficient clinical services, impede specialty development and recognition, but are priority areas for future development.