Khushi Sharma | Amrinder Kaur
| Shaveta Bhatia | Nripendra Narayan Das* | Madhulika
© 2024 The authors. This article is published by IIETA and is licensed under the CC BY 4.0 license (http://creativecommons.org/licenses/by/4.0/).
OPEN ACCESS
This study proposes an innovative approach to enhance emergency healthcare preparedness using Long Short-Term Memory (LSTM) networks, a sophisticated form of Recurrent Neural Networks (RNNs). The model leverages deep learning trends to accurately forecast healthcare requirements during crises, such as pandemics. LSTM's ability to capture long-term dependencies in data enhances predictive accuracy, making it invaluable for proactive healthcare planning. The research explores the efficacy of the LSTM-based predictive model in anticipating emergency healthcare needs, with potential implications for significantly improving response efficiency. By integrating deep learning technology, this approach represents a groundbreaking advancement, ensuring a resilient healthcare system ready to address unforeseen challenges and contribute to the overall well-being of a nation's citizens.
pandemic, healthcare, deep learning, LSTM (Long Short-Term Memory), RNN (Recurrent Neural Network)
Healthcare is the native component for the development of any country worldwide. It is the one aspect that binds the nation together in sake of the serenity of life and without it, people would have to face many challenges for existing. Being the fuel of innovation, healthcare is developing and disseminating advanced and sophisticated, life-improving therapies and at the same moment this system is striving to provide its consumers a diverse range of options.
To build a novel Health care system, delivering comprehensive health care of exceptional quality is more than a financial consideration. Before individuals have access to integrated health care, it needs time and perseverance on the part of a country.
Given the substantial healthcare, the system should make it a point to develop and innovate the models on a care system basis. Health information technology (HIT), in particular, is an important aspect of India's increasing focus on high-quality health care and efficient resource management [1].
However, growing the use of technology in healthcare will have a variety of benefits, the quality benefits will most likely be the most significant. This would, in particular, increase the chance of successful procedures and enable the supply of evidence-based decision assistance to providers, bridging the gap between evidence and practice.
Technology can have great records being clustered with healthcare in developing quality and safe environments while reducing the expenditure as well. Despite having a lot of advantages in associating Technology with healthcare, the system has still not been able to harness the power of technology and bring its assets in use.
With the recent surge in the cases of a pandemic like Covid-19, hospitals and medical institutions need to manage the patient’s influx and be decisive for the emergency admissions. It's useful to view prediction estimates, anticipate the class of new customers, have time-based predictions, and discover the variables that explain ED behavior for planning logistical needs. Finding trends in care consumption, which is accessible from the care services, is important for management. There are five components in the machine learning framework in health care is shown in Figure 1.
Figure 1. Machine learning in healthcare
The scarcity of emergency healthcare requirements is a serious issue across the country and this situation can worsen the consequences if not carefully mitigated and planned. This certainly happens due to the frequent mismatch between the resources and requirements in healthcare.
Resources controlled inside the ED or resources controlled outside the ED are certain conditions where this resource limitation can affect the actual need in emergencies. The availability of inpatient beds to accept ED patients is perhaps the most critical element affecting ED flow issues.
Artificial intelligence is advancing the world through different administrative operations. Being used for several blooming applications around the globe, its wide range of applications in healthcare marks its potential to explore the healthcare world and make it big.
Artificial Intelligence has played a major role and explored a wide-reaching potential in developing the healthcare system and breaking off the big challenge to adapt to AI in society for social good. The various assets of technology like artificial intelligence, machine learning, and deep learning are shown in Figure 2.
Figure 2. Assets of technology
The proper use of machine learning on this data has the potential to revolutionize patient risk assessment across the medical sector, particularly in infectious illnesses. As a result, tailored strategies to limit the spread of healthcare-associated infections might be developed. Healthcare is a subject of large epidemiology, requires careful processing and planning to overcome situations like a Pandemic.
Machine Learning is used to develop features and extract the calibrated parameters with different vectors. Deep Learning is thus another subset that comprehends Machine Learning in a unique form.
These networks are exciting trends in deep learning that can help in forecasting and predicting complex problems with different attributes for short-term and long-term dependencies.
The ability of LSTM to adapt and handle long-term dependencies for any computation proves to be a key factor in developing the healthcare system.
These networks overcome the backpropagation and vanishing gradient problem caused by RNNs by subsidizing cell gates used to transfigure the relevant information to make predictions.
The summary of research papers with limitations is added in Table 1.
Table 1. Summary of literature
|
Reference No. |
Motivation and Aim of the Work |
Datasets Used |
Methods Used in the Work |
Limitations |
|
[2] |
To forecast the emergency department patients for In-Hospital Mortality with an approach based on Machine learning. |
The data was used from ED’s for a healthcare system for a period of 12 months. |
A machine learning approach, a forest model was used. |
The approach was limited to specific elements of the data and leaves out unstructured data attributes. |
|
[3] |
To predict the in-patient admissions for an emergency department and execute the results in the hospitals. |
The data was collected for the ED Patients from the Boston Healthcare system. |
The model used three methods to optimize the research including. |
The research included certain data restrictions for the efficiency of the model and lacks generalized demonstration. |
|
[4] |
To analyse and depict the usage of current models to predict the hospital bed needs. |
According to the research, national data is used for prediction and then cultured into specific regions. |
This research utilized a time-series method called as Box-Jenkins approach for predicting the hospital bed occupancy. |
This approach concluded its sense with two different data driven methods for forecasting. |
|
[5] |
To forecast the requirements for emergency resources during severity and drastic situations using Natural Language Processing. |
The dataset was taken by recording the electronic health data for three different Emergency Departments for a specific time period of one year. |
The approach used for this research included deploying a LSTM model: a neural network method which was trained and tested extensively on the clinical variables. |
The research being retrospective and exclusion of a generalized form of triage instances are some limitations of the given approach. |
|
[6] |
To predict and forecast the contagious diseases using an improved deep learning approach. |
The dataset is taken from KCDC including four different kinds of data with different atributes for the prediction. |
This study used different models for prediction: LSTM model, DNN model and ARIMA model for better efficiency and improved accuracy. |
The given research used the data that was specific to using shorter periods of time and particular region specific parts. |
|
[7] |
To forecast the influenza prediction using deep learning technology LSTM. |
This study gathers details and data influenza ED visits to Taiwan Centres for Disease Control. |
This research analyses the historical data and utilizes LSTM model with the help of chronical transitory data. |
The study incorporated LSTM instead of RNN for improved efficiency. |
Machine learning aims to extract more relevant and significant information from massive amounts of data by employing a unique computational approach. The main goal of a machine learning oriented approach is to identify the variables that can contribute to an emergency situation or an infectious outbreak [8].
3.1 Deep learning
Deep learning is a subset of machine learning using the preliminary technique of building neural networks to leap forward with the processing. Artificial neural networks resemble the human brain in its structure with neuron nodes [9, 10].
This technology encompasses a variety of processing layers that do not require domain knowledge or extensive human engineering, and it comprehends with its own level of computing.
Deep learning may calibrate its internal settings by evaluating the error between the output obtained and the preferred or intended result using an objective function. The observed error can be minimized to some extent by changing these internal settings [11-13].
3.2 Recurrent Neural Networks
This network is based on the idea of storing a layer's output and feeding it back to the input in order to anticipate the layer's output [14].
Figure 3. Structure of RNN
However, training an RNN is a rather difficult method due to its vanishing gradient caused by backpropagation dealing as shown in Figure 3. RNNs are not used to deal with long-term dependencies which further causes the method to backlash [15].
It has a superior time series prediction effect. For the novel coronavirus, there is an incubation period. LSTM may be used to discover the influencing variables of probable situations when used for time series prediction [16, 17].
4.1 LSTM
When entering LSTM, the user will first travel through a forgetting gate and then via a sigmoid layer, which is also known as the forgetting gate layer.
This problem can be solved with LSTM, which can cope with sequential data while keeping the time series attribute and understanding long-term dependencies. These networks use cell gates to reduce the impact of short-term memory by subsidising the flow of important input [18].
The cell gates are nonlinear summation units that command the activation of a cell and aid in the regulation of data flow. The forget gate can be configured to multiply the cell's prior state. The input gate is in charge of selecting relevant data from the current state [19], while the output gate is in charge of determining the next concealed state. The system model for the structure of LSTM is shown in Figure 4 and Figure 5.
Figure 4. Structure of LSTM
The first component determines whether the preceding timestamp's information should be remembered or is irrelevant and can be ignored.
Figure 5. System model for prediction using LSTM
The dataset for this prediction model consists of different measures to be predicted the quantity for like number of encounters, hospital beds, medicines and other medical supplies for the given time period. The data is taken for two different countries: Austria and Belgium to predict against the parameter of Daily Hospital Occupancy. The prediction model is trained using the training dataset for the first 80 days and then tested for the training dataset for the next 20 days. The source of this data is the online portal of European Centre for Disease Prevention and Control.
Once the training phase is over, the framework uses LSTM layers to forecast the cases. The total number of instances on a given day is used as input data [20], and the model is trained for 80 days. After training the framework, the goal of this model is to forecast the number of instances over the following 20 days.
This model forecasted the hospital occupancy for the two countries: Austria and Belgium for a regular daily time interval [21, 22]. Keras is a neural network python library which works well with TensorFlow, making the creation of neural network designs easy and helps in implementing deep learning models.
The given table (Table 2) depicts the predicted number of Hospital Occupancy for the Test Data set of 20 days for the country of Austria. The model is trained using the 80 days dataset of hospital occupancy in Austria and then tested against the test dataset [23].
Table 2. Predicted and actual number of hospital occupancy for Austria
|
Sno. |
Date |
Expected Occupancy |
Actual Occupancy |
|
1 |
2020-06-20 |
63 |
66 |
|
2 |
2020-06-21 |
65 |
61 |
|
3 |
2020-06-22 |
58 |
59 |
|
4 |
2020-06-23 |
59 |
57 |
|
5 |
2020-06-24 |
53 |
59 |
|
6 |
2020-06-25 |
62 |
60 |
|
7 |
2020-06-26 |
61 |
60 |
|
8 |
2020-06-27 |
55 |
61 |
|
9 |
2020-06-28 |
59 |
62 |
|
10 |
2020-06-29 |
64 |
66 |
|
11 |
2020-06-30 |
62 |
58 |
|
12 |
2020-07-01 |
55 |
65 |
|
13 |
2020-07-02 |
57 |
62 |
|
14 |
2020-07-03 |
63 |
65 |
|
15 |
2020-07-04 |
63 |
60 |
|
16 |
2020-07-05 |
60 |
62 |
|
17 |
2020-07-06 |
63 |
67 |
|
18 |
2020-07-07 |
78 |
81 |
|
19 |
2020-07-08 |
68 |
66 |
|
20 |
2020-07-09 |
61 |
66 |
Table 2 shows the prediction accuracy of the model in respect to drawing a comparison between the predicted hospital occupancy and the actual occupancy on a daily basis for Austria.
As seen from Table 2, the model is able to forecast the occupancy on a forefront area in the right direction corresponding to the actual cases.
Table 3 demonstrate the predicted count of hospital occupancy for the test data set of 20 days for the country of Belgium. The model is trained using the 80 days dataset of hospital occupancy in Belgium and then tested against the test dataset.
Table 3 shows the prediction accuracy of the model in respect to drawing a comparison between the predicted hospital occupancy and the actual occupancy on a daily basis for Belgium.
Figure 6 shows the curves for the predicted values of Hospital Occupancy calculated by the model for the training dataset [24, 25]. The blue line is used to analyze the parameter for Belgium and the red line is used to depict the referenced attributes for Austria.
Table 3. Predicted and actual number of hospital occupancy for Belgium
|
Sno. |
Date |
Expected Occupancy |
Actual Occupancy |
|
1 |
2020-06-03 |
720 |
739 |
|
2 |
2020-06-04 |
720 |
702 |
|
3 |
2020-06-05 |
640 |
647 |
|
4 |
2020-06-06 |
577 |
573 |
|
5 |
2020-06-07 |
583 |
576 |
|
6 |
2020-06-08 |
567 |
575 |
|
7 |
2020-06-09 |
534 |
527 |
|
8 |
2020-06-10 |
483 |
484 |
|
9 |
2020-06-11 |
473 |
479 |
|
10 |
2020-06-12 |
431 |
425 |
|
11 |
2020-06-13 |
397 |
395 |
|
12 |
2020-06-14 |
404 |
400 |
|
13 |
2020-06-15 |
399 |
393 |
|
14 |
2020-06-16 |
377 |
371 |
|
15 |
2020-06-17 |
339 |
344 |
|
16 |
2020-06-18 |
325 |
340 |
|
17 |
2020-06-19 |
300 |
308 |
|
18 |
2020-06-20 |
311 |
292 |
|
19 |
2020-06-21 |
283 |
295 |
|
20 |
2020-06-22 |
297 |
293 |
Figure 6. Hospital occupancy predicted for the training set for Austria and Belgium
Figure 7. Hospital occupancy predicted for Austria
Figure 7 is used here to show and analyze the predicted results for Austria for a total span of 100 days including the 80 days of the training set and 20 days of test dataset.
Figure 8. Hospital occupancy predicted for Belgium
Figure 8 is used to analyze the predicted results for the total span of 100 days for Belgium including the training set and the test dataset.
The aim of this research is to forecast and predict the Emergency Healthcare Requirements to address the problem of emergency admissions using an improved approach of LSTM, deploying a neural network model in Deep Learning. Time series prediction can be defined well with the problem statement for the prediction of the number of emergency healthcare resources, and the LSTM model has a favorable influence.
Proposed technique is a handy way of dealing with prediction case scenario for short term dependencies, a neural network model that can act as a feedforward network and solve the issue of prediction. But due to the short term dependencies, this research has been aimed to develop a robust and flexible model using the technology of LSTM.
The network solved the problem of dealing with the long time series data using their specially calibrated cell gates which helped in subsidizing the relevant information and leaving out the rest. This technique helped the model become efficient and effective as well.
This research solved its purpose of building a flexible model to predict emergency healthcare requirements using deep learning optimizing LSTM. In the future, the model is aimed to improve its prediction accuracy reaching out to a larger set of variables and a probable dataset based on real-time data.
I would like to extend my heartfelt gratitude to my faculty guide Dr. Madhulika Bhatia for guiding me throughout as well as Father Dr. Sachin Sharma for motivation.
[1] Carroll, M., Cullen, T., Ferguson, S., Hogge, N., Horton, M., Kokesh, J. (2011). Innovation in Indian healthcare: Using health information technology to achieve health equity for American Indian and Alaska Native populations. Perspectives in Health Information Management/AHIMA, American Health Information Management Association, 8(Winter).
[2] Bates, D.W. (2002). The quality case for information technology in healthcare. BMC Medical Informatics and Decision Making, 2(1): 1-9. https://doi.org/10.1186/1472-6947-2-7
[3] Combes, C., Kadri, F., Chaabane, S. (2014). Predicting hospital length of stay using regression models: Application to emergency department. In 10ème Conférence Francophone de Modélisation, Optimisation et Simulation-MOSIM’14.
[4] Rahimian, F., Salimi-Khorshidi, G., Payberah, A.H., Tran, J., Ayala Solares, R., Raimondi, F., Nazarzadeh, M., Canoy, D., Rahimi, K. (2018). Predicting the risk of emergency admission with machine learning: Development and validation using linked electronic health records. PLoS Medicine, 15(11): e1002695. https://doi.org/10.1371/journal.pmed.1002695
[5] Peck, J.S., Benneyan, J.C., Nightingale, D.J., Gaehde, S.A. (2012). Predicting emergency department inpatient admissions to improve same-day patient flow. Academic Emergency Medicine, 19(9): E1045-E1054. https://doi.org/10.1111/j.1553-2712.2012.01435.x
[6] Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., Wang, Y., Dong, Q., Shen, H., Wang, Y. (2017). Artificial intelligence in healthcare: Past, present and future. Stroke and Vascular Neurology, 2(4). https://doi.org/10.1136/svn-2017-000101
[7] Wiens, J., Shenoy, E.S. (2018). Machine learning for healthcare: On the verge of a major shift in healthcare epidemiology. Clinical Infectious Diseases, 66(1): 149-153. https://doi.org/10.1093/cid/cix731
[8] Kai, Y., Lei, J., Yuqiang, C., Wei, X. (2013). Deep learning: Yesterday, today, and tomorrow. Journal of computer Research and Development, 50(9): 1799-1804.
[9] Verma, H., Kumar, S. (2019). An accurate missing data prediction method using LSTM based deep learning for health care. In Proceedings of the 20th International Conference on Distributed Computing and Networking, pp. 371-376. https://doi.org/10.1145/3288599.3295580
[10] Taylor, R.A., Pare, J.R., Venkatesh, A.K., Mowafi, H., Melnick, E.R., Fleischman, W., Hall, M.K. (2016). Prediction of in-hospital mortality in emergency department patients with sepsis: A local big data-driven, machine learning approach. Academic Emergency Medicine, 23(3): 269-278. https://doi.org/10.1111/acem.12876
[11] Farmer, R.D., Emami, J. (1990). Models for forecasting hospital bed requirements in the acute sector. Journal of Epidemiology & Community Health, 44(4): 307-312. https://doi.org/10.1136/jech.44.4.307
[12] Sterling, N.W., Brann, F., Patzer, R.E., Di, M., Koebbe, M., Burke, M., Schrager, J.D. (2020). Prediction of emergency department resource requirements during triage: An application of current natural language processing techniques. Journal of the American College of Emergency Physicians Open, 1(6): 1676-1683. https://doi.org/10.1002/emp2.12253
[13] Chae, S., Kwon, S., Lee, D. (2018). Predicting infectious disease using deep learning and big data. International Journal of Environmental Research and Public Health, 15(8): 1596. https://doi.org/10.3390/ijerph15081596
[14] Ramlakhan, S., Saatchi, R., Sabir, L., Singh, Y., Hughes, R., Shobayo, O., Ventour, D. (2022). Understanding and interpreting artificial intelligence, machine learning and deep learning in Emergency Medicine. Emergency Medicine Journal, 39(5): 380-385. https://doi.org/10.1136/emermed-2021-212068
[15] Boerman, A.W., Schinkel, M., Meijerink, L., van den Ende, E.S., Pladet, L.C., Scholtemeijer, M.G., Zeeuw, J., van der Zaag, A.Y., Minderhoud, T.C., Elbers, P.W.G., Wiersinga, W.J., de Jonge, R., Kramer, M.H.H., Nanayakkara, P.W. (2022). Using machine learning to predict blood culture outcomes in the emergency department: A single-centre, retrospective, observational study. BMJ Open, 12(1): e053332. https://doi.org/10.1136/bmjopen-2021-053332
[16] Pease, M., Arefan, D., Barber, J., Yuh, E., Puccio, A., Hochberger, K., Nwachuku, E., Roy, S., Casillo, S., Temkin, N., Okonkwo, D.O., Wu, S. (2022). Outcome prediction in patients with severe traumatic brain injury using deep learning from head CT scans. Radiology, 304(2): 212181. https://doi.org/10.1148/radiol.212181
[17] Saadatmand, S., Salimifard, K., Mohammadi, R., Marzban, M., Naghibzadeh-Tahami, A. (2022). Predicting the necessity of oxygen therapy in the early stage of COVID-19 using machine learning. Medical & Biological Engineering & Computing, 60(4): 957-968. https://doi.org/10.1007/s11517-022-02519-x
[18] Kijpaisalratana, N., Sanglertsinlapachai, D., Techaratsami, S., Musikatavorn, K., Saoraya, J. (2022). Machine learning algorithms for early sepsis detection in the emergency department: A retrospective study. International Journal of Medical Informatics, 160: 104689. https://doi.org/10.1016/j.ijmedinf.2022.104689
[19] Sánchez-Salmerón, R., Gómez-Urquiza, J.L., Albendín-García, L., Correa-Rodríguez, M., Martos-Cabrera, M.B., Velando-Soriano, A., Suleiman-Martos, N. (2022). Machine learning methods applied to triage in emergency services: A systematic review. International Emergency Nursing, 60: 101109. https://doi.org/10.1016/j.ienj.2021.101109
[20] Rakhra, A., Jain, I., Gupta, R., Bhatia, M. (2021). Predicting the prevalence rate of COVID-19 falsity on temperature. In 2021 11th International Conference on Cloud Computing, Data Science & Engineering (Confluence), Noida, India, pp. 327-331. https://doi.org/10.1109/Confluence51648.2021.9377198
[21] Chaudhary, R., Garg, A., Bhadauria, M. (2018). Predictive Analytics for LAMA and Absconding Behaviour of Patient. In: Satapathy, S., Bhateja, V., Das, S. (eds) Smart Computing and Informatics. Smart Innovation, Systems and Technologies, vol 78. Springer, Singapore. https://doi.org/10.1007/978-981-10-5547-8_66.
[22] Singh, A., Bhatia, M., Garg, A. (2022). Prediction of abnormal pregnancy in pregnant women with Advanced maternal age and Pregestational Diabetes using Machine learning models. In 2022 12th International Conference on Cloud Computing, Data Science & Engineering (Confluence), Noida, India, pp. 262-267. https://doi.org/10.1109/Confluence52989.2022.9734210
[23] Bhatia, M., Mittal, M. (2017). Big data & deep data: minding the challenges. In: Deep Learning for Image Processing Applications. IOS Press, Netherland, pp. 177-193.
[24] Sood, A., Hooda, M., Dhir, S., Bhatia, M. (2018). An initiative to identify depression using sentiment analysis: A machine learning approach. Indian Journal of Science and Technology, 11(4): 1-6. https://doi.org/10.17485/IJST/2018/V11I4/119594
[25] Shivam, Bareja, P., Madhulika, B. (2018). Appearance of machine and deep learning in image processing: A portrayal. International Journal of Computational Intelligence & IoT, 2(4).
