Applications of Queuing Theory and Machine Learning Techniques in Healthcare

Abstract

This thesis focuses on the application of queuing theory and machine learning techniques in healthcare management system, a field where optimization of resources is very crucial. Efficient management of hospital resources such as hospital beds, machines, equipments and staff can significantly impact both costs and patient outcomes. Having too many resources can lead to overhead expenses, while insufficient resources may result in critical situations, such as a shortage of beds during peak times. Therefore, predicting the required number of resources is important, especially when there is sudden disease outbreaks or pandemics, which pose challenges to hospital management. Queuing theory has a lots of applications in healthcare. For instance, it can optimize resource allocation in various departments, such as pharmacy units, Pediatrics, Intensive care units and emergency departments. We have utilized open Jackson queuing networks to reduce waiting times in pharmacies. Along with using open jackson queuing networks, we have also used simulation modelling and a comparison between the results obtained via the two different techniques of queuing theory and simulation has been done. Similarly, queuing models have proven effective in calculating the waiting time and other performance measures of customers in ultrasound and tomography labs. A mathematical model namely SEIRD model (Susceptible, Exposed, Infected, Recovered, Dead), which is particularly useful for studying dynamics of various pandemics, such as malaria and COVID-19. This model helps in forecasting number of patients and understanding disease spread. By incorporating such kind of mathematical models into healthcare management system, we can enhance decision-making and resource planning, ultimately improving efficiency and patient care. Next in the processing of improving healthcare services with the help of queuing theory and machine learning techniques, we focused on a pathology laboratory experiencing very high volumes of samples, leading to a considerable amount of delays in test results. This laboratory was chosen due to its extensive sample database, which allowed us to explore potential improvements. After investigation, we found that predicting turnaround time (TAT) and informing patients in advance could definitely enhance their experience. By better managing the idle time between a patient’s arrival and delivery of reports, we can optimize the process. Moreover, patients generally face decisions based on the expected waiting time for their results. If a laboratory’s turnaround time is too long, patients might opt for a faster and better alternative. Our predicted TAT can provide valuable information to help patients make more informed choices based on their priorities, such as cost versus speed of service. We utilized machine learning models to predict TAT in the laboratory setting and we iv realized that these models can also be used in distinguishing between in-care and out-care patients. We worked with a dataset that included demographic information (such as gender, age, and height) of patients and blood test results (such as blood sugar levels, body temperature, HB) to determine whether a patient requires in-care or out-care treatment, facilitating better resource allocation and patient management. In addition, we applied machine learning models to analyze the length of stay (LOS) of patients, an important factor which affects hospital resource needs, including bed availability. For this analysis, we have used a dataset covering hospital resources from different countries, which included details on the number of beds, MRI machines, and CT scanners over several years. We examined how technological advancements have influenced hospital resources and patient lenth of stay. Our findings indicated that improvements in technology have led to a reduction in patient length of stay over time. Notably this has uncovered valuable insights into the strong correlation between hospital beds and patient LOS, revealing hidden patterns that can inform future resource planning and management strategies. During our search for healthcare datasets, we encountered an important issue of maternal mortality and it remains a significant problem, particularly in developing countries and low-resource areas. Investigations in this process revealed that a lack of prper knowledge about maternal healthcare and safety measures contributes to high number of maternal and neonatal deaths. To address this, we explored a dataset containing information on blood sugar levels, blood pressure (both diastolic and systolic), body temperature, and other relevant metrics for pregnant women. We have predicted the risk levels which was categorized as low, medium, or high and identify high-risk cases early. By focusing on women at higher risk, we aimed to improve early interventions and reduce complications associated with maternal health. We have successfully concluded our thesis, summarizing the outcomes of all research conducted thus far. Additionally, we have explored the future potential of this work, discussing how the generalizability and performance of the developed models can be enhanced. This could be achieved by integrating more diverse, comprehensive, and real-time datasets in future efforts.