A machine learning–based clinical predictive tool to identify patients with a high risk of medication error

doi:10.21203/rs.3.rs-4912669/v1

Download PDF

Article

A machine learning–based clinical predictive tool to identify patients with a high risk of medication error

https://doi.org/10.21203/rs.3.rs-4912669/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

A medication error is an inadvertent failure in the drug therapy process that can cause serious harm to patients by increasing morbidity and mortality and are associated with significant economic costs to the healthcare system. Medication reconciliation is the most cost-effective intervention and can result in a 66% reduction in medication errors. To improve patient safety, we developed a machine learning-based tool that prioritizes patients at risk of medication errors upon admission to the hospital to ensure that they undergo medication reconciliation by clinical pharmacists. The data were collected from the electronic health records of patients admitted to Reims University Hospital who underwent medication reconciliation between 2017 and 2023. The data from 12,604 patients were used to train four machine learning-based models based on 52 variables in the development dataset. These models were used to prioritize admitted patients according to their likelihood of being exposed to a medication error. Our models, particularly the voting classifier model, demonstrated good performance (a recall of 0.75, precision of 0.65, F1 score of 0.70 and AUROC of 0.74). In a retrospective evaluation simulating real-life use, the voting classifier model successfully identified 45% of the total patients selected who were found to have at least one unintended discrepancy compared to 21% when using the existing tool. The positive experimental results of this tool showed a superior improvement of 113% over the existing tool by targeting patients at risk of medication errors upon admission to Reims University Hospital.

Health sciences/Health care

Health sciences/Medical research

Health sciences/Biomarkers/Predictive markers

machine learning

prediction model

medication errors

medication reconciliation

patient safety

Medication error (ME) is defined by the European Medicines Agency as an unintentional failure in the process of drug therapy that results in, or is likely to result in, harm to the patient [1]. Errors in prescribing, dispensing, drug duplication, dosing errors, storing, preparing, and administering medication are the most common preventable causes of unwanted adverse events in the practice of medication. It can occur at any time (most cases occurred on the first day of hospitalization) [2]. It can occur throughout the patient's care pathway and can result in a risk or an adverse event for the patient [3]. However, approximately half of adverse drug events are preventable medication errors [4].

There is potential for many medication errors to occur due to the lack of communication between health professionals, insufficient patient information on how to take their medicine, polymedication of the patient and the risk of confusion between his or her different treatments and inadequate information or packaging of the medicine. These MEs can cause an increase in morbidity and mortality, are associated with a significant economic cost and are estimated to cost US$42 billion annually worldwide [5]. This figure is 0.7% of all global health spending. MEs are the third leading cause of death in the United States and Europe, surpassed only by heart disease and cancer [6]. Among the various actions taken to prevent the occurrence of these MEs is medication reconciliation (MR).

MR is a clinical pharmacy approach that contributes to tackling the burden of medication discrepancies and subsequent patient harm at care transitions (admission, transfer, and discharge). MR is the most cost-effective intervention [7] and can result in a 66% reduction in medication errors [8].

Currently, MR is flourishing increasingly in France due to the favorable regulatory environment. In fact, since 2016, a decree has been issued to make clinical pharmacy a compulsory activity for internal use [9]. In addition, since January 2018, MR has been an indicator of contracts for improving the quality and efficiency of care and has been a binding contract between regional health agencies and health institutions [10]. Many studies have demonstrated the benefit of MR on patient admission [11–13]. Moreover, in France, MR at patient admission has been developed in many hospitals and, more specifically, in clinical departments such as surgery [14, 15], geriatrics [16–18], emergencies [19] and internal medicine departments [20, 21].

At the University Hospital Center (CHU) of Reims, MR is carried out retroactively upon admission to the hospital by 5th -year pharmacy students present in care services. This pharmaceutical activity has been an integral part of their mission since 2014. Students were pretrained on this activity at the College of Pharmacy and were accompanied by a reference pharmacist from the care service. MR is carried out according to a process consisting of 3 stages: actively seeking information on the patient's healthcare products, carrying out the medication review, and updating the prescription and patient records. However, MR on admission is a time-consuming and expensive process that presents a hospital-wide challenge. MR takes an average of 50 minutes per patient [22], necessitating the deployment of more clinical pharmacists, which many healthcare facilities cannot afford and are unwilling to do. Many previous studies have investigated variables that influence pharmaceutical interventions and can be used as predictors [23–25]. Other studies have discussed developing a score to identify patients at risk of medication errors using a multivariate logistic regression model [26–28], but to the best of the authors' knowledge, none have focused on machine learning-based techniques with patients admitted to different hospital wards, making this paper a novel contribution to this field.

At present, at the CHU, no instructions are given to pharmacy students performing MR regarding patient selection. Medication reconciliations are carried out on adult patients admitted to medical, surgical and obstetric wards, as well as in follow-up care and rehabilitation. These procedures were carried out within 72 hours of patient admission. Owing to the human resources allocated, it is not possible to carry out MR for all patients admitted to our institution. To increase the efficiency of this activity, we need to be able to identify those patients most at risk of unintentional discrepancies and prioritize the MR of these patients. At the CHU of Reims, we used the software Easily to computerize the care files for patients. Easily covers a wide range of functionalities, including the visualization of laboratory results, clinical files, care files and prescriptions and the management of medical letters, including digital dictation and the management of appointments. However, at present, no study has identified patients at risk of MEs upon hospitalization at the entire CHU of Reims. Currently, the increase in the number of electronic health records and the development of big data analytics are paving the way for the use of artificial intelligence (AI) and machine learning (ML) techniques to process these data, which makes MR more effective and helps reduce MEs [29]. Therefore, the aim of this study was to develop a machine learning-based tool to assist clinical pharmacists in selecting patients at high risk of medication errors for medication reconciliation. This approach could improve patient safety, lead to a more efficient process, prevent medication errors, reduce disease complications, and reduce the cost and burden on the healthcare system.

2.1 Data collection

The prediction tool developed in this study was derived based on the data of patients admitted to the CHU (Reims, France). CHU is a 2,382-bed general hospital that receives approximately 93,000 clinical admissions each year and provides both surgical and medical services. In the CHU, all patient files are typically digitized and recorded using Easily software (except for neonatology and intensive care units). Prescription orders and MRs by clinical pharmacists are managed with the same software, which includes all patient details, medical and nursing notes, laboratory results, and vital signs.

Only patients who received MR between 2017 and 2023 were considered. MR consists of a standardized medication interview with the patient conducted by a practitioner (pharmacist or pharmacist-in-training). The data were extracted directly from the Easily software database. In addition, because MR sheets are often poorly linked to the respective hospitalizations in Easily software, relinking was performed using patient identifiers, hospitalization dates, and MR dates (considered reliable information). MRs that could not be associated with hospitalization were excluded. The final cohort was validated in consultation with pharmacists. Throughout this study, all the data were stored on a secure local server, and pseudonymous patient identities were used instead of their names (an identification number for each patient).

2.2 Data preprocessing

Data preprocessing is the first step in the ML algorithm and is used to transform the raw data into useful, understandable and efficient formats. Out of 350,540 patients admitted to the hospital between 2017 and 2023, only 12,604 patients were randomly selected for MR, with a total of 52 variables representing each patient's record.

In this study, the first step in data preprocessing was to improve the data quality, mainly by rectifying outlier values and by filling in possible empty values. To this end, each incorrect or empty numerical value (excluding counters) was filled with the mean of the associated column, while the counter and Boolean values were filled with zeros. In addition, checks on uniqueness (particularly at the level of identifying variables) and correct ordering (particularly at the level of dates, before conversion into numeric variables) were carried out. The second step was data normalization, which was used to adjust the values of variables in the dataset to a common scale. Here, min–max normalization was used where all variable values ranged between [0,1].

The final dataset used included 4361 negative patients and 2839 positive patients. To determine whether handling balanced data contributes to improved performance, training datasets were prepared with/without resampling. The preprocessed dataset was undersampled by the edited nearest neighbor (ENN) method [30] and then oversampled by adaptive synthetic sampling (ADASYN).[31] The ENN and ADASYN algorithms were applied using the imbalanced-learn library [32] in the Python 3.9 language with the default settings.

Table 1

Baseline patient characteristics
Sociodemographic	State at the moment of the reconciliation:
	- Age - Sex - Body mass index (BMI) - Patient came from his home - Patient came from an institution
Hospitalization	State at the moment of the admission:
	- Admission during vacation period - Admission during week-end - Admission during preshift (5:30 pm to 6 pm or 8 am and 8:30 am) - Admission during night shift (6 pm to 8 am) - Admission from emergency unit - Admission in emergency unit within the past 30 days - Previous hospitalization within the past 30 days - Previous hospitalization within the past 6 months - First hospitalization - Number of doctors associated with the patient
	State at the moment of the reconciliation:
	- Occupancy rate of the ward - Hospitalization in surgical unit - Hospitalization in rehabilitation unit - Number of hours elapsed since the admission - Number of hours elapsed between the admission and the first prescription
Clinical	State at the moment of the reconciliation:
	- Speech disorder was detected - No speech disorders - Hearing disorder was detected - No hearing disorder - Visual disturbance was detected - No visual disturbance - Cognitive impairment was detected - No cognitive impairment - Swallow disorder was detected - No swallow disorder - Patient had a nasogastric tube or stoma - Patient did not have a nasogastric tube or stoma - Charlson score
Medication	State of the last valid medication prescription at the moment of the reconciliation:
	- Number of prescribed drugs per ATC drug class,[33] 14 variables were used to represent classes (A, B, C, D, G, H, J, L, M, N, P, R, S, and V) - Number of prescribed drugs without ATC classification - Prescriber was identified and was an intern - Prescriber was identified and was not an intern
Reconciliation	State at the moment of the admission:
	- Number of days elapsed between the admission and the last medication reconciliation for the patient - Total number of reconciliations for the patient

To develop a comprehensive prediction tool for identifying patients at risk of medication errors, a total of 52 variables (Table 1) were extracted directly or indirectly (by calculation) from the patients’ records. These variables were chosen on the basis of the advice of the pharmacists and according to similar studies on medication reconciliation [2, 21, 25].

2.3 Prediction tool

Four machine learning algorithms—logistic regression (LR), support vector machine (SVM), extreme gradient boosting (XGB), and voting classifier (VC)—were used to develop the prediction models. Python 3.9.6 was used for coding the algorithm, and the scikit-learn library [34] was used for all the ML models, except for XGB, for which the xgboost library [35] was used. VC is a machine learning model that trains on a set of base models and predicts an output (class) based on hard voting or soft voting criteria. Hard voting aggregates the outputs of each base model and predicts the class label with the highest number of votes, while soft voting aggregates the probabilities of each class label and predicts the class label with the highest average probability. In this study, the SVM, LR, XGB, and random forest models were used as base classifiers for the VC classifier with hard voting criteria. Here, the default settings of the machine learning algorithms were used.

2.4 Validation

The proportion of patients with MR data in the database was used as the gold standard, to which the outcomes predicted by the prediction model were compared. To address the problem of overfitting and to determine whether the model is generalizable, internal validation was performed using fivefold stratified cross-validation for each prediction model.

To evaluate model performance, several widely recognized metrics, F1 score, precision, and recall, were selected to measure model performance for predicting patients at risk of medication errors. The F1 score is defined as the harmonic mean of precision and recall. Precision measures the proportion of examples classified as positive that are truly positive, whereas recall measures the proportion of actual positives that are identified correctly. The precision is defined by precision = tp/(tp + fp), and the recall (also known as sensitivity) is defined by recall = tp/(tp + fn), where tp, fp, and fn are the numbers of true positives, false positives, and false negatives, respectively. Further metrics for the statistical performance analysis included the area under the receiver operating characteristic curve (AUROC). The receiver operating characteristic curve describes the trade-off between the true positive rate (TPR), also known as the sensitivity/recall, and the false positive rate (FPR) at various threshold settings. The AUROC is a measure of model performance: the closer the value is to 1, the better the prediction performance. We also used the area under the precision‒recall curve (AUCPR), which provides a more accurate representation of the impact the model can have on the pharmacist’s work [36], given that the dataset is imbalanced. The AUCPR describes the trade-off between the positive predictive value (also known as precision) and the TPR at various threshold settings.

Table 2

Comparison of the performances of different ML models on imbalanced datasets
	Imbalanced dataset
Metric	LR	SVM	XGB	VC
Recall	0.59	0.44	0.48	0.67
Precision	0.53	0.58	0.60	0.55
F1 score	0.56	0.50	0.54	0.60
AUROC	0.68	0.68	0.70	0.71
AUCPR	0.60	0.60	0.64	0.65

Table 3

Comparison of the performances of different ML models on a balanced dataset (training)
	Balanced dataset
Metric	LR	SVM	XGB	VC
Recall	0.63	0.66	0.55	0.75
Precision	0.52	0.51	0.58	0.52
F1 score	0.57	0.58	0.56	0.61
AUROC	0.68	0.68	0.71	0.71
AUCPR	0.60	0.59	0.64	0.64

Table 4

Comparison of the performances of different ML models on balanced datasets (training and validation)
	Balanced dataset
Metric	LR	SVM	XGB	VC
Recall	0.63	0.66	0.55	0.75
Precision	0.65	0.65	0.69	0.65
F1 score	0.64	0.65	0.61	0.70
AUROC	0.70	0.70	0.72	0.74
AUCPR	0.70	0.71	0.73	0.75

Over a period of 7 years, data were collected only for patients who were hospitalized at various hospital departments for at least 24 hours (average age 73 years, male/female, 47%/53%). During this period, 12,604 patients were randomly selected to undergo medication reconciliations. Pharmacist intervention was recommended for 2,878 of the 12,604 patients, meaning that 22.8% of patients were at risk of at least one unintended discrepancy.

The results of the different prediction models are shown in Tables 2, 3, and 4. Table 2 contains the results for the prediction models (LR, SVM, XGB, and VC) based on the original imbalanced dataset. Table 3 contains the corresponding results when the training dataset is resampled (under/oversampled by the ENN and ADASYN), Table 4 contains the corresponding results when the training dataset is resampled (under and oversampled by the ENN and ADASYN) and the validation dataset is undersampled.

As with most healthcare datasets, one of the major hurdles when learning from imbalanced data is that the minority class is usually the class of interest. In the training set, having a class with few examples would cause any model to be skewed toward improper classification, as it will not receive enough examples for the class of interest. Therefore, balancing the dataset helps prevent the model from being biased toward the dominant class.

Visual inspection of the values in Tables 2–4 enables comparisons between the effectiveness of different prediction models. From the analysis of the different metrics used, it can be concluded that resampling the dataset leads to a viable improvement in prediction performance. The results in Tables 3 and 4 show that the prediction models (LR, SVM, XGB, and VC) produced higher recall, precision, F1 score, AUROC, and AUCPR values when the balanced dataset was used than when the imbalanced dataset was used, as shown in Table 2. As shown in Table 3, the VC model exhibited the highest performance among the four ML models and had the highest recall, F1 score, AUROC, and AUCPR values. Although the VC model showed higher recall (75%), the precision (52%) was not as good as the recall. The low precision and high recall indicate that the class was well detected but also included observations of other classes (false positives), which can be confirmed by the results of the AUCROC curve in Fig. 1. Table 4 shows the results of the prediction models when balanced training and validation datasets were used (which means fewer false positives). Table 4 shows a significant improvement in precision for all the models, with the best results (75% recall, 65% precision, 70% F1 score, 74% AUROC, and 75% AUCPR) obtained by the VC model.

The main aim of this study was to improve patient safety by developing a tool that prioritizes admitted patients at risk of medication errors to undergo medication reconciliation by clinical pharmacists. To determine the extent to which the objectives underlying this study have been met, first, we will discuss the existing tool used at the CHU of Reims to select patients admitted for medication reconciliation and the extent of its effectiveness. We will later discuss the results obtained using our developed tool and compare them with the existing tool used to determine how efficient our tool is.

At our CHU of Reims, medication reconciliation is performed on a random sample of patients who have been hospitalized for at least 24 hours and have had at least one prescription order. Here, we will discuss the results of the existing tool used at our CHU of Reims for selected patients for medication reconciliation in 2022. Out of 47,876 patients admitted to the hospital in 2022, only 2,037 (4.3%) were randomly selected for medication reconciliation. We found that medication reconciliation detected at least one unintended discrepancy in only 26% of the selected patients (536 of 2037 patients), leading to pharmacist intervention. These results can be read from two different points of view. On the one hand, the current procedure has succeeded in intercepting 26% of the positive patients who require pharmaceutical intervention. On the other hand, the presence of 26% of the patients within the selected random sample was considered high, which indicated that a large number of patients were not selected and may need pharmacist intervention. Due to limited resources and capabilities, the number of patients selected for medication reconciliation did not exceed 4.3% of the total patients admitted. Therefore, the presence of 74% of negative cases is considered a waste of the already limited resources that were supposed to be used to examine the files of other patients who may be at risk of medication errors.

In what follows, we will discuss the results of our ML tools and, in particular, the VC model, as it achieved the best results. The results in Table 2 show that the performance of the ML tools was affected by class imbalance, in which examples in training data belonging to one class greatly outnumber the examples in the other class. Two methods were used to balance the training dataset, allying a known undersampling method (ENN) and the oversampling method (ADASYN), to produce better defined class clusters. Our comparative results in Tables 2 and 3 show that using prediction tools with a balanced dataset provides more accurate results than does using an imbalanced dataset considering the recall metric.

Due to the imbalance of the dataset used for validation, the AUROC can provide a false impression of the quality of the classifier (see Table 2). As shown in Table 2, the AUROC was high, although the recall was not high, suggesting that the classifier performed well, which was not the case. However, although most of the negative cases are correctly classified, many of the positives may be misclassified, which is most important for model performance. We also noticed that (Table 3) despite the significant improvement in the recall metric, improvements in other metrics were not noticeable, especially in terms of precision, AUROC and AUCPR. The low precision is due to the increase in false positives, which can be confirmed by examining the AUROC curve in Fig. 1. The ROC curve plots the TPR versus FPR at different classification thresholds, and as the threshold decreases, the recall increases because we identify more patients who have medication errors. However, as our recall increases, our precision decreases because, in addition to increasing the number of true positives, we increase the number of false positives. Therefore, the AUCPR and precision were also used to evaluate the quality of the ML tools. Thus, the recall and AUCPR values in Table 2 show that the performance of the ML model is not good, contrary to what the AUROC value reveals. To confirm the above results, we balanced the validation dataset using the undersampled ENN method such that it contained an equal number of positives and negatives, which improved all the metrics used (see Table 4). The results in Table 4 clearly revealed that the improvement in all the metrics was associated with an increase in the number of false positives because of the imbalance of the validation dataset. Figure 2 also shows significant improvement in the AUCPR compared to the AUCPR, as shown in Fig. 1.

Although the above results are very promising compared to those of existing tools, we still need to test our ML tool in a real-life evaluation to confirm its effectiveness. Therefore, to validate our tool, we performed a retrospective evaluation simulating real-life use to confirm its effectiveness. To conduct this evaluation, data from 317 patients who underwent medication reconciliation during the period 11/27/2023 to 02/04/2024 were used. Of the 317 patients, 93 were found to have at least one unintended discrepancy, which led to pharmacist intervention. Our ML tool was used to prioritize the 317 patients in descending order according to their likelihood of experiencing a medication error. A specific number of patient files (110 files) with high scores were subsequently selected. On the other hand, a group of clinical pharmacists was also recruited to carry out a selection process for the same number of patient files (110 files) but randomly. Finally, for each approach, the total number of patients who had at least one unintended discrepancy was recorded.

To determine the extent to which the objectives behind this study were met, first, we will discuss the results of the random approach used to select patients admitted for medication reconciliation and its effectiveness. Then, we will discuss the results obtained using our ML tool for the same number of patients and compare them with those of the random approach. Figure 3 shows the results for both approaches. Using a random approach, 110 patients were randomly selected for medication reconciliation. Twenty-three of the 110 patients had at least one unintended discrepancy, while 87 patients did not. In other words, the random approach succeeded in identifying only 21% of the total selected patients. When using the ML tool, 110 patients were selected for medication reconciliation, and 45% of the total selected patients were found to have at least one unintended discrepancy, which led to pharmacist intervention, with an improvement of 113% compared to the random approach. These results confirm that the ML tool outperformed the existing tool used at CHU of Reims in terms of its ability to identify a greater number of patients exposed to medication errors. This approach could lead to increased capabilities of the clinical pharmacy department so that the medication reconciliation process would be more efficient and would reduce the burden on clinical pharmacists.

In summary, improving patient safety by reducing medication errors is a top priority in hospital health systems. Currently, medication reconciliation by clinical pharmacists is the standard method for preventing potential adverse drug events. These interventions are often limited by human resources, and patients at high risk for errors in their prescription orders need to be targeted. A strength of this study is that it included patients from internal medicine, surgical, and obstetric wards, as well as follow-up care and rehabilitation units; therefore, the findings are representative of daily clinical practice.

The single-center nature of our study was one of the limitations of this study. In addition, neonatology and intensive care unit patients were not included because these units do not use Easily software for medication prescription. Thus, there is no evidence of the accuracy of our model in identifying patients at high risk in these units; therefore, our model cannot be generalized to these patients.

It would be interesting in future work to improve the performance of our tool. The performance of an ML tool can be enhanced by obtaining abundant and accurate datasets. However, adding new and accurate data requires carrying out a large number of medication reviews, which is limited by a lack of well-trained human resources. This has motivated us to deploy our tool throughout other hospitals, as this will enable us to benefit from the expertise of a larger number of clinical pharmacists to confirm our findings and as an automated way to produce data that are used to update the model's learning dataset.

We developed and internally validated a machine learning-based predictive tool to predict patients with a high risk of at least one medication error during admission. Our predictive tool works as a digital tool to improve patient safety by identifying as many patients at risk of medication error as possible using available human resources. The positive experimental results of this tool showed a superior improvement of 113% over the existing tool by targeting patients at risk of medication errors upon admission to Reims University Hospital.

Author Contribution

AA and LG: Data curation, designing and implementing the software, investigating, and experimenting. AA, LG, TV, JC, PQ, VV and RM: Conception and interpretation of the data. AA: Writing the original manuscript. VV and RM: Resources, supervision and project administration. All the authors reviewed the manuscript critically for scientific content, and all the authors gave final approval of the manuscript for publication.

Competing interests

The authors declare no competing interests.

Funding

The authors did not receive support from any organization for the submitted work.

Additional Information

Approval: The Institut d'Intelligence Artificielle en Santé (IIAS) Research Ethics Committee has confirmed

Accordance: We confirmed that all experiments in this study were performed in accordance with the relevant guidelines and regulations.

Informed consent: the informed consent was waived by the IIAS research.

Availability of Data

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Medication errors | European Medicines Agency. https://www.ema.europa.eu/en/human-regulatory-overview/post-authorisation/pharmacovigilance-post-authorisation/medication-errors (accessed 28 May 2024)
Nguyen T-L, Leguelinel-Blache G, Kinowski J-M, et al. Improving medication safety: Development and impact of a multivariate model-based strategy to target high-risk patients. PLOS ONE. 2017;12:e0171995. doi: 10.1371/journal.pone.0171995
Qu’est ce qu’une erreur médicamenteuse ? - ANSM : Agence nationale de sécurité du médicament et des produits de santé. https://archive.ansm.sante.fr/Declarer-un-effet-indesirable/Erreurs-medicamenteuses/Qu-est-ce-qu-une-erreur-medicamenteuse/(offset)/0 (accessed 23 August 2023)
Santé AFA. Tous impliqués pour signaler les événements indésirables associés aux soins. Fr. Assos Santé. 2022. https://www.france-assos-sante.org/2022/11/18/tous-impliques-pour-signaler-les-evenements-indesirables-associes-aux-soins/ (accessed 23 August 2023)
Donaldson LJ, Kelley ET, Dhingra-Kumar N, et al. Medication Without Harm: WHO’s Third Global Patient Safety Challenge. The Lancet. 2017;389:1680–1. doi: 10.1016/S0140-6736(17)31047-4
Gøtzsche PC. Our prescription drugs kill us in large numbers. Pol Arch Med Wewn. 2014;124 11:628–34.
Mekonnen AB, McLachlan AJ, Brien JE. Pharmacy-led medication reconciliation programmes at hospital transitions: a systematic review and meta-analysis. J Clin Pharm Ther. 2016;41:128–44. doi: 10.1111/jcpt.12364
Karnon J, Campbell F, Czoski-Murray C. Model-based cost-effectiveness analysis of interventions aimed at preventing medication error at hospital admission (medicines reconciliation). J Eval Clin Pract. 2009;15:299–306. doi: 10.1111/j.1365-2753.2008.01000.x
Décret n° 2019-489 du 21 mai 2019 relatif aux pharmacies à usage intérieur. 2019.
https://www.omedit-grand-est.ars.sante.fr/media/83094/download?inline (Accessed 23 August 2023)
Mueller SK, Sponsler KC, Kripalani S, et al. Hospital-Based Medication Reconciliation Practices: A Systematic Review. Arch Intern Med. 2012;172:1057–69. doi: 10.1001/archinternmed.2012.2246
Anderson LJ, Schnipper JL, Nuckols TK, et al. Effect of medication reconciliation interventions on outcomes: A systematic overview of systematic reviews. Am J Health Syst Pharm. 2019;76:2028–40. doi: 10.1093/ajhp/zxz236
Deawjaroen K, Sillabutra J, Poolsup N, et al. Clinical usefulness of prediction tools to identify adult hospitalized patients at risk of drug-related problems: A systematic review of clinical prediction models and risk assessment tools. Br J Clin Pharmacol. 2022;88:1613–29. doi: 10.1111/bcp.15104
Duval M, Satori D, Al Wazzan N, et al. Conciliation médicamenteuse à l’entrée des patients dans une unité de chirurgie orthopédique : retour d’expérience sur plus de 2 ans de pratique. Ann Pharm Fr. 2021;79:700–9. doi: 10.1016/j.pharma.2021.03.003
Vallecillo T, Slimano F, Moussouni M, et al. Development and validation of a ready-to-use score to prioritise medication reconciliation at patient admission in an orthopaedic and trauma department. Eur J Hosp Pharm. 2022;29:264–70. doi: 10.1136/ejhpharm-2020-002283
Boissinot L, Bachalat N, Perrier-Cornet E, et al. Conciliation des traitements médicamenteux en gériatrie : Pertinence et faisabilité. Pharmactuel. 2014;47.
Saint-Germain P, Ruelle M, Mary A, et al. Impact clinique des divergences de traitement constatées chez 200 patients conciliés dans un service de gériatrie aiguë. Rev Médecine Interne. 2016;37:667–73. doi: 10.1016/j.revmed.2016.02.017
Barbier G, Berthou J, Koeberle S, et al. Sécurisation du parcours patient en gériatrie : quels facteurs de risque d’erreurs médicamenteuses à l’admission ? Pharm Hosp Clin. 2020;55:275–81. doi: 10.1016/j.phclin.2020.04.011
T. Poinsat JC M Sitbon, O Ganansia, C Gerlier, Y Bézie. Conciliation médicamenteuse d’entrée en amont des services d’aval des urgences : faisabilité et intérêts. Pharm Hosp Clin. 2020;7260:227–320. doi: http://dx.doi.org/10.1016/j.phclin.2020.11.004
Harang C, Martin S, Belmenouar O, et al. Conciliation médicamenteuse à l’admission : un réel impact sur la qualité de notre prise en charge. Rev Médecine Interne. 2018;39:A202. doi: 10.1016/j.revmed.2018.10.192
Audurier Y, Roubille C, Manna F, et al. Development and validation of a score to assess risk of medication errors detected during medication reconciliation process at admission in internal medicine unit: SCOREM study. Int J Clin Pract. 2020;75:e13663. doi: 10.1111/ijcp.13663
Barnsteiner JH. Medication Reconciliation. In: Hughes RG, ed. Patient Safety and Quality: An Evidence-Based Handbook for Nurses. Rockville (MD): Agency for Healthcare Research and Quality (US) 2008.
Cornuault L, Mouchel V, Phan Thi T-T, et al. Identification of variables influencing pharmaceutical interventions to improve medication review efficiency. Int J Clin Pharm. 2018;40:1175–9. doi: 10.1007/s11096-018-0668-y
Hias J, Van der Linden L, Spriet I, et al. Predictors for unintentional medication reconciliation discrepancies in preadmission medication: a systematic review. Eur J Clin Pharmacol. 2017;73:1355–77. doi: 10.1007/s00228-017-2308-1
Vallecillo T, Slimano F, Moussouni M, et al. Development and validation of a ready-to-use score to prioritise medication reconciliation at patient admission in an orthopaedic and trauma department. Eur J Hosp Pharm. 2022;29:264–70. doi: 10.1136/ejhpharm-2020-002283
Falconer N, Liow D, Zeng I, et al. Validation of the assessment of risk tool: patient prioritisation technology for clinical pharmacist interventions. Eur J Hosp Pharm. 2017;24:320. doi: 10.1136/ejhpharm-2016-001165
Ebbens MM, Laar SA van, Wesselink EJ, et al. Prospective Validation of a Risk Prediction Model to Identify High-Risk Patients for Medication Errors at Hospital Admission. Ann Pharmacother. 2018;52:1211–7. doi: 10.1177/1060028018784905
Bosma LBE, Rein N van, Hunfeld NGM, et al. Development of a multivariable prediction model for identification of patients at risk for medication transfer errors at ICU discharge. PLOS ONE. 2019;14:e0215459. doi: 10.1371/journal.pone.0215459
Jiang F, Jiang Y, Zhi H, et al. Artificial intelligence in healthcare: past, present and future. Stroke Vasc Neurol. 2017;2. doi: 10.1136/svn-2017-000101
Batista GEAPA, Prati RC, Monard MC. A study of the behavior of several methods for balancing machine learning training data. ACM SIGKDD Explor Newsl. 2004;6:20–9. doi: 10.1145/1007730.1007735
He H, Bai Y, Garcia EA, et al. ADASYN: Adaptive synthetic sampling approach for imbalanced learning. 2008 IEEE International Joint Conference on Neural Networks (IEEE World Congress on Computational Intelligence). 2008:1322–8.
Lemaître G, Nogueira F, Aridas CK. Imbalanced-learn: A Python Toolbox to Tackle the Curse of Imbalanced Datasets in Machine Learning. J Mach Learn Res. 2017;18:1–5.
ATCDDD - ATC/DDD Index. https://atcddd.fhi.no/atc_ddd_index/ (accessed 14 June 2024)
Pedregosa F, Varoquaux G, Gramfort A, et al. Scikit-learn: Machine Learning in Python. J Mach Learn Res. 2011;12:2825–30.
Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. San Francisco California USA: ACM 2016:785–94.
Davis J, Goadrich M. The relationship between Precision-Recall and ROC curves. Proceedings of the 23rd international conference on Machine learning. New York, NY, USA: Association for Computing Machinery 2006:233–40.

No competing interests reported.

Download PDF

Editorial decision: Revision requested
07 Oct, 2024
Reviews received at journal
19 Sep, 2024
Reviews received at journal
17 Sep, 2024
Reviewers agreed at journal
09 Sep, 2024
Reviewers agreed at journal
08 Sep, 2024
Reviewers invited by journal
08 Sep, 2024
Editor assigned by journal
08 Sep, 2024
Editor invited by journal
29 Aug, 2024
Submission checks completed at journal
28 Aug, 2024
First submitted to journal
14 Aug, 2024

You are reading this latest preprint version

A machine learning–based clinical predictive tool to identify patients with a high risk of medication error

Status:

Version 1

Abstract

Figures

1 Introduction

2 Materials and Methods

2.1 Data collection

2.2 Data preprocessing

2.3 Prediction tool

2.4 Validation

3 Results

4 Discussion

5 Conclusion

Declarations

References

Additional Declarations

Status:

Version 1