Figure - available from: BMC Medical Informatics and Decision Making
This content is subject to copyright. Terms and conditions apply.
Dispensing pattern of simvastatin to atorvastatin switching and their respective PDC measures. A Lipid-lowering PDC is calculated by summing their respective PDC and setting an upper bound of 100
Source publication
Background
Chronic conditions place a considerable burden on modern healthcare systems. Within New Zealand and worldwide cardiovascular disease (CVD) affects a significant proportion of the population and it is the leading cause of death. Like other chronic diseases, the course of cardiovascular disease is usually prolonged and its management neces...
Citations
... To facilitate earlier intervention, a sequential model capable of making predictions at any given time step would be markedly more beneficial. While some subsequent studies have introduced sequential models (Hsu et al., 2022;Singh et al., 2022;Schleicher et al., 2023), their scope was restricted to predicting adherence alone. Our study enhances this approach by incorporating a state-action model, which can predict both adherence and score/state. ...
... A sequential model that can make predictions at any specific time step would significantly enhance the ability for early intervention. Hsu et al. (2022) investigated the advantages of incorporating patient history into the prediction of medication adherence. They assessed the performance of temporal neural network models, particularly LSTM and simple recurrent neural networks, and compared these with non-temporal neural networks, ridge classifiers, and logistic regression. ...
Objective
Subcutaneous Immunotherapy (SCIT) is the long-lasting causal treatment of allergic rhinitis (AR). How to enhance the adherence of patients to maximize the benefit of allergen immunotherapy (AIT) plays a crucial role in the management of AIT. This study aims to leverage novel machine learning models to precisely predict the risk of non-adherence of AR patients and related local symptom scores in 3 years SCIT.
Methods
The research develops and analyzes two models, sequential latent-variable model (SLVM) of Stochastic Latent Actor-Critic (SLAC) and Long Short-Term Memory (LSTM). SLVM is a probabilistic model that captures the dynamics of patient adherence, while LSTM is a type of recurrent neural network designed to handle time-series data by maintaining long-term dependencies. These models were evaluated based on scoring and adherence prediction capabilities.
Results
Excluding the biased samples at the first time step, the predictive adherence accuracy of the SLAC models is from 60% to 72%, and for LSTM models, it is 66%–84%, varying according to the time steps. The range of Root Mean Square Error (RMSE) for SLAC models is between 0.93 and 2.22, while for LSTM models it is between 1.09 and 1.77. Notably, these RMSEs are significantly lower than the random prediction error of 4.55.
Conclusion
We creatively apply sequential models in the long-term management of SCIT with promising accuracy in the prediction of SCIT nonadherence in AR patients. While LSTM outperforms SLAC in adherence prediction, SLAC excels in score prediction for patients undergoing SCIT for AR. The state-action-based SLAC adds flexibility, presenting a novel and effective approach for managing long-term AIT.
... Therefore, DLR is a great candidate as this model constructs the trajectory by dynamically estimating adherence using previous estimations and including the autocorrelation effect. The literature also uses different machine learning approaches to model adherence for multiple periods, such as temporal deep learning for five years into the future [26] and random forests for the next two weeks [33]. Similarly, our model can predict how patient adherence will evolve, giving healthcare providers additional information beyond a classification method. ...
Long-term adherence to medication is a critical factor in preventing chronic diseases, such as cardiovascular disease. To address poor adherence, physicians may recommend adherence-improving interventions; however, such interventions are costly and limited in their availability. Knowing which patients will stop adhering helps distribute the available resources more effectively. We developed a binary integer program (BIP) model to select patients for adherence-improving intervention under budget constraints. We further studied a long-term adherence prediction model using a dynamic logistic regression (DLR) model that uses patients' claim data, medical health factors, demographics, and monitoring frequencies to predict the risk of future non-adherence. We trained and tested our predictive model to longitudinal data for cardiovascular disease in a large cohort of patients taking medication for cholesterol control seen in the national Veterans Affairs health system. Our study shows the importance of including past adherence to increase prediction accuracy. Finally, we assess the potential benefits of using the prediction model by proposing an algorithm that combines the DLR and BIP models to decrease the number of CVD events in a population.
... Therefore, DLR is a great candidate as this model constructs the trajectory by dynamically estimating adherence using previous estimations and including the autocorrelation effect. The literature also uses different machine learning approaches to model adherence for multiple periods, such as temporal deep learning for five years into the future [26] and random forests for the next two weeks [33]. Similarly, our model can predict how patient adherence will evolve, giving healthcare providers additional information beyond a classification method. ...
Long-term adherence to medication is a critical factor in preventing chronic diseases, such as cardiovascular disease. To address poor adherence, physicians may recommend adherence-improving interventions; however, such interventions are costly and limited in their availability. Knowing which patients will stop adhering helps distribute the available resources more effectively. We developed a binary integer program (BIP) model to select patients for adherence-improving intervention under budget constraints. We further studied a long-term adherence prediction model using dynamic logistic regression (DLR) model that uses patients' claim data, medical health factors, demographics, and monitoring frequencies to predict the risk of future non-adherence. We trained and tested our predictive model to longitudinal data for cardiovascular disease in a large cohort of patients taking medication for cholesterol control seen in the national Veterans Affairs health system. Our study shows the importance of including past adherence to increase prediction accuracy. Finally, we assess the potential benefits of using the prediction model by proposing an algorithm that combines the DLR and BIP models to decrease the number of CVD events in a population.
... In a similar way, our findings, which focused on time series demand prediction of medicines, can be supported in the same context as similar findings were reported in other studies conducted in different contexts or using different dataset types. For instance, Hsu et al. 2022, published a study that focused on the prediction of adherence to medical treatment with temporal modelling in cardiovascular disease management. The study highlighted that temporal models that use sequential data outperform non-temporal models, with LSTM showing better performance in prediction and achieving an area under the curve (AUC) of 0.805 [3]. ...
... The study's findings paved the way for future research by suggesting an investigation of the comparison of deep learning techniques and tree-based ensembles in various prediction instances. The findings of a such study may inform decision-making in sales management, product flow, operational and other logistic aspects, as well as promotional tactics with the goal of strengthening supply chain management [27]. Because medicines and health products are usually so costly, it is paramount to point out, in relation to the findings of our study, that the use of data-driven prediction models is a foregone decision to recommend in the health supply chain as this will optimize the accuracy of demand prediction for selected medicines. ...
The ever-accelerating revolution along with digitalization of the healthcare industry has revealed the power of machine learning and deep learning prediction models in addressing health supply chain logistic issues. The purpose of this study was to predict the demand for medicines using autoregressive integrated moving average (ARIMA) and long short-term memory (LSTM) time series models while comparatively analysing their performance for medicine demand prediction to optimize the flow of supplies in the health system. Using data generated in Rwanda public health supply chain, in our study focused on predicting the demand of the top five medicines, identified as highly supplied (amoxicillin, penicillin v, ibuprofen, paracetamol, and metronidazole). We evaluated the models’ outputs by root mean square error (RMSE) and the coefficient of determination, R-squared (R2). In comparison to ARIMA, the deep learning LSTM model revealed superior performance with better accuracy and lower error rates in predicting the demand for medicines. Our results revealed that the LSTM model has an RMSE value of 2.0 for the training set and 2.043 for the test set, with R2 values of 0.952 and 0.912, respectively. ARIMA has an RMSE value of 9.35 for the training set and 8.926 for the test set as well as R2 value of 0.24 and 0.16 for the training and test sets, respectively. Based on these findings, we recommend that the LSTM time series model should be used for demand prediction in the management of medicines and their flow within health supply chain due to its remarkable performance for prediction task when applied to the dataset of our study.
... Memory values are modified by the input gate. As the forget gate determines what details are to be discarded from a block, the output gate determines what parts of a cell's state should be output [38]. The formulation of LSTM can be given in the following formulas 1 to 6. ...
... Sequential models for adherence prediction. Hsu et al. (2022) investigated the advantages of incorporating patient history into the prediction of medication adherence. They assessed the performance of temporal deep learning models, particularly LSTM and simple Recurrent Neural Networks (RNN), and compared these with non-temporal models such as Multilayer Perceptrons (MLP), ridge classifiers, and logistic regression. ...
... To facilitate earlier intervention, a sequential model capable of making predictions at any given time step would be markedly more beneficial. While some subsequent studies have introduced sequential models (Hsu et al., 2022;Singh et al., 2022;Schleicher et al., 2023), their scope was restricted to predicting adherence alone. Our study enhances this approach by incorporating a state-action model, which can predict both adherence and score/state. ...
Objective: Subcutaneous Immunotherapy (SCIT) is the long-lasting causal treatment of allergic rhinitis. How to enhance the adherence of patients to maximize the benefit of allergen immunotherapy (AIT) plays a crucial role in the management of AIT. This study aims to leverage novel machine learning models to precisely predict the risk of non-adherence of patients and related systematic symptom scores, to provide a novel approach in the management of long-term AIT. Methods: The research develops and analyzes two models, Sequential Latent Actor-Critic (SLAC) and Long Short-Term Memory (LSTM), evaluating them based on scoring and adherence prediction capabilities. Results: Excluding the biased samples at the first time step, the predictive adherence accuracy of the SLAC models is from 60 % to 72%, and for LSTM models, it is 66 % to 84 %, varying according to the time steps. The range of Root Mean Square Error (RMSE) for SLAC models is between 0.93 and 2.22, while for LSTM models it is between 1.09 and 1.77. Notably, these RMSEs are significantly lower than the random prediction error of 4.55. Conclusion: We creatively apply sequential models in the long-term management of SCIT with promising accuracy in the prediction of SCIT nonadherence in Allergic Rhinitis (AR) patients. While LSTM outperforms SLAC in adherence prediction, SLAC excels in score prediction for patients undergoing SCIT for AR. The state-action-based SLAC adds flexibility, presenting a novel and effective approach for managing long-term AIT.
... 18 Formally, MPR is defined as the sum of the days' supplies of medication during a period divided by the number of the period. 19 While the level of optimal adherence may differ for different clinical conditions, persistency is usually defined as adherence (or MPR) of ≥80%. 20 ...
Hypertension is a chronic disease that requires long-term follow-up in many patients, however, optimal visit intervals are not well-established. This study aimed to evaluate the incidences of major cardiovascular events (MACEs) according to visit intervals. We analyzed data from 9894 hypertensive patients in the Korean Hypertension Cohort, which enrolled and followed up 11,043 patients for over 10 years. Participants were classified into five groups based on their median visit intervals (MVIs) during the 4-year period and MACEs were compared among the groups. The patients were divided into clinically relevant MVIs of one (1013; 10%), two (1299; 13%), three (2732; 28%), four (2355; 24%), and six months (2515; 25%). The median follow-up period was 5 years (range: 1745 ± 293 days). The longer visit interval groups did not have an increased cumulative incidence of MACE (12.9%, 11.8%, 6.7%, 5.9%, and 4%, respectively). In the Cox proportional hazards model, those in the longer MVI group had a smaller hazard ratio (HR) for MACEs or all-cause death: 1.77 (95% confidence interval [CI], 1.45-2.17), 1.7 (95% CI: 1.41-2.05), 0.90 (95% CI: 0.74-1.09) and 0.64 (95% CI: 0.52-0.79), respectively (Reference MVI group of 75-104 days). In conclusion, a follow-up visits with a longer interval of 3-6 months was not associated with an increased risk of MACE or all-cause death in hypertensive patients. Therefore, once medication adjustment is stabilized, a longer interval of 3-6 months is reasonable, reducing medical expenses without increasing the risk of cardiovascular outcomes.
Aim:
To develop deep learning models to predict nursing need proxies among hospitalised patients and compare their predictive efficacy to that of a traditional regression model.
Design:
This methodological study employed a cross-sectional secondary data analysis.
Methods:
This study used de-identified electronic health records data from 20,855 adult patients aged 20 years or older, admitted to the general wards at a tertiary hospital. The models utilised patient information covering the preceding 2 days, comprising vital signs, biomarkers and demographic data. To create nursing need proxies, we identified the six highest-workload nursing tasks. We structured the collected data sequentially to facilitate processing via recurrent neural network (RNN) and long short-term memory (LSTM) algorithms. The STROBE checklist for cross-sectional studies was used for reporting.
Results:
Both the RNN and LSTM predicted nursing need proxies more effectively than the traditional regression model. However, upon testing the models using a sample case dataset, we observed a notable reduction in prediction accuracy during periods marked by rapid change.
Conclusions:
The RNN and LSTM, which enhanced predictive performance for nursing needs, were developed using iterative learning processes. The RNN and LSTM demonstrated predictive capabilities superior to the traditional multiple regression model for nursing need proxies.
Implications for the profession:
Applying these predictive models in clinical settings where medical care complexity and diversity are increasing could substantially mitigate the uncertainties inherent in decision-making processes.
Patient or public contribution:
We used de-identified electronic health record data of 20,855 adult patients about vital signs, biomarkers and nursing activities.
Reporting method:
The authors state that they have adhered to relevant EQUATOR guidelines: STROBE statement for cross-sectional studies.
Impact:
Despite widespread adoption of deep learning algorithms in various industries, their application in nursing administration for workload distribution and staffing adequacy remains limited. This study amalgamated deep learning technology to develop a predictive model to proactively forecast nursing need proxies. Our study demonstrates that both the RNN and LSTM models outperform a traditional regression model in predicting nursing need proxies. The proactive application of deep learning methods for nursing need prediction could help facilitate timely detection of changes in patient nursing demands, enabling the effective and safe nursing services.
