Out of the 22 selected papers [8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29], 17 focus on predicting the duration of surgical planning [8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24]. This finding underscores the crucial role of accurate estimation in surgical case duration for effective operating room management. It presents a complex and multifaceted challenge that profoundly impacts OR scheduling, resource allocation, and overall operational efficiency. Our previous review [4] primarily highlighted the promising results of a proprietary algorithm known as leap Rail® [31]. While it exhibited an improvement in predictive accuracy compared to traditional methods, our updated review reveals a more nuanced picture. More recent studies, such as the work by Bartek and colleagues [8], have delved deeper into the use of machine learning models, emphasizing the importance of surgeon-specific models. These newer models outperform service-specific ones and significantly enhance the accuracy of case-time predictions, offering substantial benefits in terms of operating room management. Our updated analysis also demonstrates the dominance of XGBoost in machine learning models over other algorithms, including the random forest model and linear regression. XGBoost’s superior predictive capabilities are showcased, which is a notable deviation from the earlier review’s focus on leap Rail® [30]. This underlines the rapid advancements in machine learning technology and its potential to refine surgical case duration predictions. However, is important to keep in mind that different outcomes could require different ML algorithms. [32] Another key finding in the previous review was the potential cost savings associated with accurate surgical case duration predictions in robotic surgery. However, our updated review provides new insights. Jiao and colleagues [11] introduced the use of modular artificial neural networks (MANN) for predicting remaining surgical duration. MANNs are neural networks equipped with external memory. They excel at tasks requiring context and sequential reasoning, making them suitable for certain clinical applications. They leveraged anesthesia records from a diverse range of surgical populations and hospital types, showcasing the robustness and adaptability of their model. MANN consistently outperformed Bayesian statistical approaches, particularly during the last quartile of surgery, indicating its potential for cost savings and operational efficiency improvements. The study also assessed the generalizability and transferability of the MANN model. It found that even healthcare systems with lower operative volumes could benefit from fine-tuning a model trained at larger nearby systems. It also highlighted the lack of meaningful information in the anesthesia record during certain phases of surgery, suggesting room for improvement. This study underscores the rapid advances in machine learning algorithms and their application in real-world surgical scenarios. Variational autoencoders (VAEs), which are generative models designed for learning latent representations of data, also fit in this context. They consist of an encoder and a decoder. The encoder maps input data to a probability distribution in a latent space, and the decoder reconstructs data from samples in this latent space. Linking advanced models like MANNs and VAEs to clinical sense implies that these models could contribute to the field of personalized medicine by learning patient-specific representations, enabling tailored treatment plans and also address clinical needs, enhance diagnostics, improve patient outcomes, or streamline healthcare processes [33]. The work conducted by Strömblad et al. [23], a single-center, randomized clinical trial brought additional insights. They explored the accuracy of predicting surgical case durations using a machine learning model in comparison to the existing scheduling-flow system. This research emphasized the benefits of a comprehensive and data-driven prediction approach, which resulted in a significant reduction in mean absolute error (MAE), contributing to enhanced prediction accuracy. Importantly, this decrease in MAE translated into reduced patient wait times without adversely affecting surgeon wait times or operational efficiency, indicating a harmonious balance between efficiency and patient outcomes. This study is the first and only randomized clinical trial on the subject, to our knowledge, representing a significant milestone.

When comparing the reviews, both the previous [4] and the updated one underscore the potential benefits of improved prediction accuracy in surgical scheduling and operating room management. However, the newer studies provide more specific insights into practical implications. Bartek and colleagues’ work [8] shows a reduction in wait times and resource utilization through the implementation of machine learning-driven models. This has a significant impact on patient outcomes without disrupting operational efficiency, reinforcing the value of these predictive models in real-world healthcare settings. In comparing the updated review of predictive models for PACU length of stay with the previous version [4], we can discern a substantial evolution also in this field. The earlier review had already acknowledged the importance of improving hospital organization and internal logistics to reduce the costs associated with time and space waste in healthcare [4]. It had highlighted issues of congestion in the PACU due to inadequate surgical planning, which often led to patients being held in the OR when PACU beds were unavailable, incurring higher costs. In the current update, we have expanded our analysis to include more recent studies, specifically focusing on predicting PACU length of stay, and their findings are striking. One study conducted by Schulz and colleagues [25] utilized a dataset of 100,511 cases to develop predictive models for PACU length of stay. They considered variables such as patient age, surgical urgency, duration of surgery, and more to create a neural network model. Notably, the study evaluated individual anesthesiologists, categorizing them based on their mean PACU length of stay. The predictive model, relying on routinely collected administrative data, significantly explained variations in individual anesthesiologists’ mean PACU length of stay. This study underscored the practicality of deploying predictive models within existing hospital infrastructure. Tully and colleagues’ research [27], another notable study in this field, aimed to develop a model that could classify patients at high risk for a prolonged PACU stay of ≥ 3 h. The study considered factors like surgical procedure, patient age, and scheduled case duration. The most effective model was XGBoost, which significantly improved the ability to predict prolonged PACU stays. Furthermore, by using the XGBoost model’s predictions, cases were re-sequenced based on the likelihood of a prolonged PACU stay, which led to a substantial reduction in the number of patients in the PACU after hours. These recent studies collectively signify a remarkable shift in the field of PACU length of stay prediction. They highlight the potential of predictive models, machine learning, and data-driven approaches to enhance healthcare quality and operational efficiency. The adoption of big data analytics and optimization of case sequencing have clear implications for improving patient outcomes and resource allocation. It is evident that these models hold significant promise for healthcare institutions, potentially offering considerable cost savings and enhanced patient care. When comparing these recent findings with the previous version of the review, we see a marked advancement in the sophistication of predictive models. The earlier version primarily emphasized the issue of inefficient PACU use and its financial implications, highlighting the potential for cost savings through improved surgical planning. The new studies demonstrate not only the cost-saving potential but also the power of data-driven predictive models, which can significantly enhance the efficiency and effectiveness of healthcare operations.

One of the significant challenges in the healthcare industry is the unexpected cancellation of surgical cases. Surgical cancellations not only disrupt the workflow of healthcare facilities but also pose risks to patient safety and satisfaction [34]. To address this issue and optimize surgical scheduling, ML techniques have emerged as a promising solution for the early detection of potential cancellations. Comparing the updated review with the previous version [4] reveals substantial advancements in this critical aspect of healthcare management. In the earlier review [4], the focus was on the high costs associated with surgical case cancellations, particularly highlighting the cost variation across different types of surgeries. It underscored the need for automatic classification methods to detect high-risk cancellations from large datasets. Furthermore, the review discussed the potential for ML algorithms, specifically random forest, in identifying surgeries at high risk of cancellation, with the promise of optimizing healthcare resource utilization and cost-efficiency. The current review continues to emphasize the significance of addressing surgical case cancellations in healthcare. For example, Luo et al. [28] significantly contribute to the field by leveraging ML to identify high-risk cancellations. Their research focuses on a dataset of elective urologic surgeries, comprising over 5,000 cases, with the aim of identifying surgeries prone to cancellation due to institutional resource- and capacity-related factors. Authors employed three ML algorithms, including random forest, support vector machine, and XGBoost, and evaluated their performance across various metrics. Their findings revealed the suitability of ML models for identifying surgeries at low risk of cancellation, effectively narrowing down the pool of surgeries with higher risk. Moreover, the random forest models displayed good efficacy in distinguishing high-risk surgeries, with an area under the curve (AUC) exceeding 0.6, indicating an interesting result in this context. Different sampling methods allowed for adjustments in model performance, highlighting the trade-offs between sensitivity and specificity. The study concluded that ML models are feasible for identifying surgeries at risk of cancellation. In a subsequent study by Zhang and colleagues [29] from the same center, the focus shifted to providing effective methodologies for recognizing high-risk surgeries prone to cancellation. They also utilized the same dataset but explored a variety of machine learning models, including random forest, logistic regression, XGBoost, support vector machine, and neural networks. The study identified the random forest model as the top-performing algorithm, achieving a high accuracy of 0.8578 and an AUC of 0.7199. Despite the high specificity and negative predictive value, the study acknowledged the need for improving sensitivity and positive predictive value in identifying high-risk cases. In summary, both studies [28, 29] aim to address the challenge of surgical case cancellations in healthcare using machine learning techniques. They highlight the importance of selecting the right machine learning algorithm for this task and acknowledge the need for improving sensitivity and positive predictive value. Both studies [28, 29] acknowledge limitations related to their focus on elective urologic surgeries within a single hospital and suggest the potential for future research to expand to diverse healthcare settings for improved generalizability. Comparing the two reviews, the earlier version [4] emphasized the need for ML algorithms to address surgical case cancellations but did not delve into specific research findings for a lack of studies on the argument. In contrast, in the current version we provide in-depth insights into the suitability of different ML models for identifying high-risk surgeries. Both reviews share a common theme: the critical role of ML techniques in addressing surgical case cancellations to enhance healthcare resource utilization and cost-efficiency.

In summary, the comparison between the two editions of the systematic reviews on the artificial intelligence integration in operative room management highlights a remarkable evolution in each domain. In the case of surgical case duration estimation, the newer review showcases a shift towards machine learning-based models, notably XGBoost, and a heightened focus on surgeon-specific models. This means the realization of machine learning’s potential, promising increased precision in predictions, cost reduction, and enhanced operating room management. Similarly, in the PACU length of stay prediction domain, the updated review underscores the transformative potential of predictive models, emphasizing the value of big data analytics, optimized case sequencing, and risk-adjusted metrics for improving patient outcomes and resource allocation. It acknowledges the challenges of real-world implementation and the need for further validation through prospective studies and collaborative efforts. Overall, the updated review provides deeper insights into the practical applications of these advanced techniques, offering healthcare providers and managers valuable tools to enhance efficiency, reduce costs, and improve patient care. The shift towards center-specific models in healthcare, particularly for organizational aspects, merits in-depth exploration. This trend reflects the growing recognition that customization based on center-specific variables, such as the type of surgeon or anesthetist, can lead to more accurate predictions and better resource allocation. The balance between clinical and organizational applications in these models remains a key consideration. While clinical models focus on patient-specific factors, organizational models, including center-specific ones, primarily address resource optimization, scheduling efficiency, and cost reduction. The choice between center-specific and clinical models ultimately depends on the specific goals and priorities of a healthcare institution. Regarding clinical implementation, it is crucial to investigate how many of these advanced models will progress beyond research to practical application. The shift towards real-world usability is gaining traction, but not all studies provide tools or software for direct application. A critical aspect is the integration of these models into daily work routines. Successful implementation often involves interdisciplinary collaboration between data scientists, healthcare professionals, and administrators. These tools can be used by a range of stakeholders, including surgeons, anesthetists, scheduling teams, and hospital administrators. Different outputs from these models serve varied purposes. For example, clinical models can guide treatment decisions, while organizational models can enhance resource allocation and scheduling efficiency. The extent to which these models are designed for easy integration and use in daily healthcare operations is a key area of investigation, ultimately impacting their practical utility and impact on patient care and healthcare management.