1 Introduction

This section introduces the research background and significance of telemedicine, automated classification and prioritisation of remote patients. Moreover, the academic literature of related works is presented and criticised before describing the research contribution of the study. Five sequential questions are presented as follows.

First question: ‘What is telemedicine and why it’s important?’.

Telemedicine, which is currently a global trend, aims to bring health services closer to patients. This technology can be applied to several situations; it allows the provision of health services at home or during emergency cases (Ahmed et al. 2020). Telemedicine attends to different patient needs to provide healthcare services by offering inexpensive medical services in underserved and remote areas (Mohammed et al. 2021a, b). Remote health monitoring is an important issue in telemedicine. Hence, various network technologies and wireless communications have been developed to provide healthcare services anywhere at any time (Kalid et al. 2018a, b; Qiao and Koutsakis 2008). ‘Remote patients’ are those patients who are far from a hospital and utilise telemedicine applications (Baig and Gholamhosseini 2013; Okura et al. 2016). The severity of remote patients is often addressed by classifying them into different triage levels.

Second question: ‘What is the importance of remote patient classification?’.

Classification is a method of categorising patients during emergency situations and determining the order by which aid is provided on the basis of disease severity and urgency (Tomozawa et al. 2009). Automation of this method may effectively improve the provision of quality healthcare services and hospitalisation, thereby saving lives (Lee et al. 2018). The growth of an ageing population and an increase in the frequency of natural disasters can overwhelm the capacity of healthcare systems and the number of already insufficient specialists (Burke et al. 2012). Triage systems can guarantee that patients in need of urgent care are promptly attended to when emergency departments (EDs) are being overwhelmed (Eijk et al. 2015). Triage systems help physicians prioritise patients according to medical urgency (Murphy et al. 2013). Rapid dissemination of the vital signs of patients is crucial to ensure reliable ED data and immediate emergency care (Zvikhachevskaya et al. 2009). Prioritisation is essential to provide prompt healthcare services. Physicians measure the vital signs of patients to determine disease severity and current health condition, prioritise patients who need immediate treatment and then provide appropriate life-saving treatment. Subsequently, they will identify, label and follow up their patients (Rodriguez et al. 2014). In local and ED settings, patients are prioritised on the basis of the criticality of their condition (Xiong et al. 2012).

Third question: ‘What is the importance of the prioritisation of remote patients?’.

During an emergency, the highest priority is given to patients needing the most urgent care in a telemedicine environment (Kalid et al. 2018a, b). Differences in the severity and urgency of patients with chronic heart diseases (CHDs) contribute to the difficulty in deciding who should be treated first among remote patients (Sarkar and Sinha 2014). Triage determines the severity of health condition in emergency settings (Moreno et al. 2016). Prioritisation is important to deliver patient records quickly and reliably, especially in emergency situations (Algaet et al. 2014). In reality, first come, first served (FCFS) care is adopted in EDs (Claudio and Okudan 2009). However, FCFS cannot be used in actual situations; hence, a rapid, well-informed and timely decision in patient prioritisation is needed (Claudio and Okudan 2009; Tan 2013). Improper classification and patient prioritisation can lead to incorrect strategic decisions which can endanger patient lives (Kalid et al. 2018a, b). In a medical situation, if the classification process is individually performed, then patient prioritisation within each category cannot be determined. Such a case will cause a decline in prioritisation performance. If patients only receive treatment with classification but without prioritisation, then the most urgent cases will be at risk. Prioritisation performance is defined as the ability to prioritise patients into categories according to different issues, namely, scalability of support, targeted tier, environment, method of prioritisation (category/order), weighting of features, ranking of multiple patient triage criteria, accuracy of patient prioritisation and handling large amounts of data (Kalid et al. 2018a, b).

Fourth question: ‘What is the criticism and gap analysis of the related academic literature?’.

To the best of our knowledge, only three studies on telemedicine environment proposed a solution for classifying and prioritising patients with CHD. In (Salman et al. 2014), a multi-source data fusion (MSHA) framework is proposed. This framework considers multiple clinical parameters as measured by wireless body area network devices, such as electrocardiogram (ECG), oxygen saturation (SpO2) and blood pressure (BP), and regards texts inputs as the health complaint. MSHA is utilised to enhance healthcare scalability challenges by improving distant classification in a telemedicine environment. In (Albahri et al. 2019a, b), patients with CHD are categorised into four emergency levels by using a new four-level remote triage and a localisation algoritm. Such algoritm is developed within a smart real-time hospital selection framework. This framework can be used to select the best healthcare provider for each patient with CHD by using integrated multi-criteria decision making (MCDM) techniques. In (Albahri et al. 2019a, b), a risk-level localisation triage within a fault-tolerant mHealth framework is developed for triaging and classifying high-risk patients with CHD and then selecting the best hospital for each patient. Only two relevant telemedicine studies proposed a prioritisation solution for patients with CHD. In (Kalid et al. 2018a, b), large data of patients with CHD were evaluated and scored according to the hybridisation of models of multi-layer analytic hierarchy process (MLAHP) and technique for order performance by similarity to ideal solution methods (TOPSIS) to improve prioritisation in a telemedicine environment. In (Salman et al. 2017), large data of patients with CHD requiring immediate medical attention were evaluated and prioritised. Prioritisation was also based on TOPSIS. However, the models of (Albahri et al. 2019a, b; Salman et al. 2014) only classify/triage emergency levels into different categories and do not prioritise the ranking within each category. Moreover, the method for evaluating the diagnostic value of patient classification presented in Kalid et al. (2018a, b) and Salman et al. (2017) is considered an inaccurate approach. These earlier studies only focused on prioritisation and neglected classification. For example, their methods assign high priority to sick patients over high-risk patients. Thus, their approaches do not simultaneously optimise all emergency levels within all categories. An approach combining the classification and prioritisation of patients with CHD within an integrated framework has not been developed in telemedicine studies. Such a combination can address the limitations of earlier approaches.

Fifth question: ‘What is the contribution of the present study?’.

A medical practitioner must categorise patients at different levels according to the condition of patients and individually prioritise them within each category to increase prioritisation performance. This approach is the primary contribution of the present study. This process is called prioritisation of emergency for emergency and urgent for urgent. The present research focuses on classifying and prioritising remote patients who seek medical services in a telemedicine environment. A framework which integrates classification and prioritisation can classify and prioritise numerous patients with CHD to receive emergency and treatment-based services.

This paper is organised as follows. Section 2 describes the research methodology. Section 3 presents the results and discussion. Section 4 reports the validation and evaluation results of the proposed framework. Section 5 concludes the study and offers research avenues for future works.

2 Methodology

This section provides an overview and explanation of the proposed methodology. Figure 1 summarises the focus of the subsequent sections.

Fig. 1
figure 1

Proposed methodology

Full size image

2.1 Identification of targeted tier in telemedicine architecture

The telemedicine architecture consists of three tiers (Zaidan et al. 2020). The client side is represented by Tiers 1 and 2, and the server side is represented by Tier 3 (Mohammed et al. 2020b). Tier 1 (data collection) data are collected from medical sensors and manual inputs (i.e. texts) related to CHDs (Albahri et al. 2018a, b, c). Three biomedical sensors are responsible for transferring the vital signs and reliable datasets (i.e. ECG, SpO2 and BP) of patients for processing by the server side (Tier 3). Text inputs include chest pain, shortness of breath, palpitation and the physical condition (i.e. rest or exercise) of patients, which are answerable by ‘yes’ (if abnormal) or ‘no’ (if normal) (Mohammed et al. 2019). Tier 2 (i.e. laptop or smartphone) is responsible for transferring the medical sensor/source signals from Tier 1 to Tier 3 by using mobile cellular networks or other communication protocols (Albahri et al. 2018a, b, c). In the server, the data from the sensors are analysed by the proposed framework. In this context, Tier 3 is the targeted tier in this study. The process at the server side is an estimation of the medical situation of patients, thereby triaging and prioritising urgent cases as defined by specific features (Albahri and Zaidan et al. 2018). The client side processes (i.e. Tiers 1 and 2) are beyond the scope of this study. Tier 3 includes real-time remote health monitoring which enables doctors to analyse data and provide patients with compatible services. The server usually contains medical records, reports on user history and database (Touati and Tabish 2013). The server is where processes and decisions are made, and it can resolve several problems. Among these issues and as stated in this research, the following requirements in the server side should be fulfilled:

  • Classification: Classify patients according to emergency level and separate them into five categories, namely, at high risk, requires urgent care, sick, cold state and normal.

  • Prioritisation: Prioritise patients in each category according to emergency status and then assign them in a queue.

2.2 Identification of patients with CHD and dataset

This section specifies the type and number of patients. The patients identified in this research (n = 500) were patients with CHD who seek medical services remotely via telemedicine. In January 2021, a data row was derived from the most reliable and relevant medical database (Goldberger et al. 2000) which currently has 9645 citations according to Google scholar index. This reliable medical database might be useful for other researchers because it contains many datasets validated and verified by medical experts. Several recent telemedicine studies have used these data in various ideas even though they were published in 2000. The first telemedicine study (Salman et al. 2014) used and processed these data for classifying patients with CHD into different emergency levels remotely. Moreover, this dataset was adopted in the prioritisation of single CHD patients (Kalid et al. 2018a, b; Salman et al. 2017) and multiple CHD patients (Mohammed et al. 2020a; Zaidan et al. 2020) in a telemedicine environment. Meanwhile, authors in (Albahri et al. 2019a, b; Albahri 2021) used the mentioned data in the classification of CHD patients followed by selecting an appropriate healthcare provider for each. The study by Salman et al. (2020) used these data for reducing the waiting time of remote patients with CHD, taking into account admitted patients in the ED. In the present study, the adopted dataset is presented in Table A.1 in Appendix. The dataset includes physical and medical information, such as gender, age, patient location and medical history in the hospital server. Males accounted for 60% of the dataset used, whereas females accounted for the remaining 40%. Furthermore, 50% of the patients were between the ages of 40 and 65 years, 40% were over the age of 65 years and 10% were under the age of 40 years.

2.3 Framework development

In this section, the classification and prioritisation framework was developed using three sequential models. The first classification model is based on Dempster–Shafer theory (Salman et al. 2014). The two other prioritisation models are based on hybridised MLAHP and TOPSIS models (Kalid et al. 2018a, b). The present study combines these classification and prioritisation models to develop a unified classification and prioritisation framework at Tier 3, as presented in Fig. 2.

Fig. 2
figure 2

Intelligent telemedicine classification and prioritisation framework

Full size image

As mentioned in Sect. 2.1, this study focused on and contributed to the medical central server (Tier 3) via two sequential and successful stages. In the first stage, the patients were classified according to case severity and separated into five categories. This process may serve as a guideline for describing emergency status and determining patients who are at high risk, require urgent situation, sick, in a cold state or normal. In the second stage, each emergency category from the first stage is prioritised according to the hybridised MCDM models, namely, MLAHP and TOPSIS. At this stage, MLAHP is used to obtain weights for each of the four sources and related features from six experts on CHD. TOPSIS is then used to prioritise and rank the available patients within each emergency level, and each patient is prioritised according to weights extracted from MLAHP. The following subsection further describes each model.

2.3.1 Classification model based on Dempster–Shafer theory

Dempster–Shafer theory is adopted at Tier 3 to classify 500 patients with CHD according to emergency status and separate them into five categories, namely, at high risk (represented in red), requires urgent care (represented in orange), sick (represented in yellow), in a cold state (represented in blue) and normal (represented in green). The classification process is based on ECG sensors, BP sensors, SpO2 sensors and texts. The combination of multiple sources within the data fusion framework improves the estimation of the current vital status of remote users and generates a triage tag called ‘Priority Code (PC)’ (Salman et al. 2014). PC is utilised to classify patients and identify each emergency case. A number from 0 to 100 is then used to draw a conclusion. This number denotes the PC and can be considered as a guideline for describing the emergency status, as shown in Table 1 (Salman et al. 2014). For example, if the PC is equal to 75, then the patient is classified as a high-risk patient (red) and thus should be afforded with appropriate medical care. This process is considered a description guideline in addressing patients who are at high risk, require urgent care, sick, in a cold state and normal.

Table 1 Medical classification guideline
Full size table

2.3.2 Prioritisation models based on MLAHP and TOPSIS

At this step, the patients’ classification according to emergency status is presented in the classification model. The patients are categorised into five triage levels. Each triage level has a corresponding emergency level. MCDM is used to decide which patient is prioritised for each triage level. The integrated MLAHP and TOPSIS prioritises patients according to a set of measurements, as shown in Fig. 3 (Kalid et al. 2018a, b). In this context, prioritisation is achieved on the basis of the MLAHP and TOPSIS models.

Fig. 3
figure 3

Integrated MLAHP and TOPSIS model for patient prioritisation (Kalid et al. 2018a, b)

Full size image

The MLAHP model is utilised to distribute weights for each feature in the criterion hierarchy (Khatari et al. 2021; Malik et al. 2021; Mohammed et al. 2021a, b; Sharma et al. 2020). Each main feature is rated in the hierarchy for each patient with CHD involved in the evaluation. By comparison, the MLAHP model is used to obtain ratio scales for each criterion used in the evaluation process (Kalid et al. 2018a, b). At this stage, several steps are performed to assign adequate weights to multi-source criteria by using MLAHP (Prasad et al. 2020). A cardiologist designs and distributes a comparison questionnaire to six experts on heart diseases (Kalid et al. 2018a, b). The experts first render their judgement on the basis of four primary criteria and related characteristics, namely, SpO2, ECG, BP and text (sources), to compare them and demonstrate the relative importance of each criterion (Kalid et al. 2018a, b).

The TOPSIS model is used during this stage (Kalid et al. 2018a, b) to prioritise patients with CHD. TOPSIS can prioritise and rank patients according to emergency status and show them in a queue at each classification level. The overall weights of MLAHP are drawn to address the main weaknesses of TOPSIS, which are its lack of provision for weight generation and inconsistent checks on judgements (Kalid et al. 2018a, b).

The available alternatives are scored in a descending order, and patients requiring the most urgent care are prioritised according to TOPSIS. The aggregate score provides an idea of which patients should be given more urgent attention (Ramasamy et al. 2020). As with other ranking options, relying on people to rank the most urgent case is always possible. On the basis of geometric distance from negative and positive ideal solutions, TOPSIS assigns the rank to each patient at each classification level. According to this technique, patients who require urgent care in emergency settings would have the shortest geometric distance to the ideal positive solution but the longest geometric distance to the ideal negative solution (i.e. have the highest value amongst all patients) (Kalid et al. 2018a, b). The steps of the TOPSIS method (Kalid et al. 2018a, b) are described as follows:

Step 1::

Construct the normalised decision matrix

This process attempts to transform the dimensions of various attributes (vital features) into non-dimensional attributes. Step 1 also allows the attributes to be compared (Alaa et al. 2019; Zaidan et al. 2020). The matrix (\({x}_{ij}\))_(m*n) is normalised from (\({x}_{ij}\))_(m*n) to the matrix R = (\({r}_{ij}\))_(m*n) via the normalisation method shown in Eq. (1):

$${{r}_{ij}=x}_{ij}/\sqrt{\sum _{i=1}^{m}{x}_{ij}^{2}}.$$
(1)

This process produces a new matrix R:

$$\begin{array}{c}\\R=\left[\begin{array}{cc}\begin{array}{cc}{r}_{11}& {r}_{12}\\ {r}_{21}& {r}_{22}\end{array}& \begin{array}{cc}\dots & {r}_{1n}\\ \dots & {r}_{2n}\end{array}\\ \begin{array}{cc}\vdots & \vdots \\ {r}_{m1}& {r}_{m2}\end{array}& \begin{array}{cc}\vdots & \vdots \\ \dots & {r}_{mn}\end{array}\end{array}\right].\end{array}$$
(2)
Step 2::

Construct the weighted (scoring points) and normalised decision matrixes

In this process, the weights for each attribute are calculated according to the MLAHP model, a set of weights \(w={w}_{1},{w}_{2},{w}_{3},\cdots,{w}_{j},\cdots,{w}_{n}\) from the decision-maker is accommodated to the normalised decision matrix; the resulting matrix can be calculated by multiplying each column of the normalised decision matrix R with its associated weight \({\mathrm{w}}_{\mathrm{j}}\) (Kalid et al. 2018a, b). The set of the weights is equal to 1, as illustrated in Eq. (3):

$$\sum _{j=1}^{m}{w}_{j}=1.$$
(3)

This process produces a new matrix V:

$$V=\left[\begin{array}{cc}\begin{array}{cc}{v}_{11}& {v}_{12}\\ {v}_{21}& {v}_{22}\end{array}& \begin{array}{cc}\dots & {v}_{1n}\\ \dots & {v}_{2n}\end{array}\\ \begin{array}{cc}\vdots & \vdots \\ {v}_{m1}& {v}_{m2}\end{array}& \begin{array}{cc}\vdots & \vdots \\ \dots & {v}_{mn}\end{array}\end{array}\right] =\left[\begin{array}{cc}\begin{array}{cc}{{w}_{1}r}_{11}& {w}_{2}{r}_{12}\\ {{w}_{1}r}_{21}& {w}_{2}{r}_{22}\end{array}& \begin{array}{cc}\dots & {w}_{n}{r}_{1n}\\ \dots & {w}_{n}{r}_{2n}\end{array}\\ \begin{array}{cc}\vdots & \vdots \\ {w}_{1}{r}_{m1}& {w}_{2}{r}_{m2}\end{array}& \begin{array}{cc}\vdots & \vdots \\ \dots & {{w}_{n}r}_{mn}\end{array}\end{array}\right].$$
(4)
Step 3::

Determine the ideal and negative ideal solutions

In this process, two artificial alternatives, namely, A* (ideal alternative) and A − (negative ideal alternative), are defined by Eqs. (5) and (6), respectively:

$${A}^{\mathrm{*}}=\left\{\left(\left(\underset{i}{\mathrm{max}}{ v}_{ij}|j\in J\right), \left(\underset{i}{\mathrm{min}}{v}_{ij}|j\in {J}^{-}\right)|i=\mathrm{1,2},\dots, m\right)\right\}=\left\{{v}_{1}^{\mathrm{*}}, {v}_{2}^{\mathrm{*}},\dots , {v}_{j}^{\mathrm{*}},\cdots {v}_{n}^{\mathrm{*}}\right\},$$
(5)
$${A}^{-}=\left\{\left(\left(\underset{i}{\mathrm{min}}{v}_{ij}|j\in J\right),\left(\underset{i}{\mathrm{max}}{v}_{ij}|j\in {J}^{-}\right)|i=\mathrm{1,2},\dots, m\right)\right\}=\left\{{v}_{1}^{-},{v}_{2}^{-},\dots, {v}_{j}^{-},\cdots {v}_{n}^{-}\right\},$$
(6)

where \(J\) is a subset of\(\left\{\mathrm{i}=\mathrm{1,2},\dots ,\mathrm{m}\right\}\), which presents the benefit attribute (i.e. offering an increasing utility with its higher values) and \({J}^{-}\) is the complement set of\(J\). The opposite could be added as well for the cost type attribute as denoted by \({J}^{c}\).

Step 4::

Calculate the separation measurement by using the Euclidean distance

A separation measurement is performed by calculating the distance between each alternative in V and the ideal vector A* with the Euclidean distance, which is given by Eq. (7) as follows:

$${S}_{{i}^{\mathrm{*}}}=\sqrt{\sum _{j=1}^{n}{\left({v}_{ij}-{v}_{j}^{\mathrm{*}}\right)}^{2}}, i=\left(\mathrm{1,2},\cdots m\right).$$
(7)

Similarly, the separation measurement for each alternative in V from the negative ideal A− is given by Eq. (8) as follows:

$${S}_{{i}^{-}}=\sqrt{\sum _{j=1}^{n}{\left({v}_{ij}-{v}_{j}^{-}\right)}^{2}},$$
(8)

where \(i=\left(\mathrm{1,2},\cdots m\right).\)

At the end of step 4, the values of S_(i*) and S_(i−) for each alternative are counted. These two values represent the distance between each alternative and both the ideal and negative ideal solutions.

Step 5::

Calculate the closeness to the ideal solution

The closeness of A_i to the ideal solution A* is defined in Eq. (9):

$${C}_{{i}^{\mathrm{*}}}={S}_{{i}^{-}}/\left({S}_{{i}^{-}}+{S}_{{i}^{\mathrm{*}}}\right), 0<{C}_{{i}^{\mathrm{*}}}<1, i=\left(\mathrm{1,2},\cdots m\right),$$
(9)

where C_(i*) = 1, if and only if (A_i = A*). Similarly, C_(i*) = 0, if and only if (A_i = A−).

Step 6::

Prioritise patients according to their closeness to the ideal solution

The set of patients〖A〗_i can now be prioritised in the descending order of 〖C〗_ (i*). The highest value indicates the optimal performance.

3 Results and discussion

This section presents the results of patient classification and prioritisation. As described in Sect. 2.3, three models were used to develop the classification and prioritisation framework. One of them is related to classification, whereas the two other models are related to prioritisation.

3.1 Results of patient classification

The proposed framework classified patients into different emergency levels as mentioned in Sect. 2.3.1. The output of Dempster–Shafer theory is PC. Table 2 presents the classification results of 500 patients with CHD.

Table 2 Patient classification results
Full size table

Five rules were applied to classify patients: 66 were high-risk patients, 151 patients required urgent care, 260 patients were sick, 23 patients were in a cold state and no patient was normal. Emergency status was identified and classified. The structure of the classification results is shown in Fig. 4. All patients within each level are then prioritised.

Fig. 4
figure 4

Results of classification process

Full size image

3.2 Results of patient prioritisation

The hybridised MLAHP and TOPSIS decision-making model was used to score and prioritise patients in each level. At this stage, the classification results of 500 patients from Sect. 3.1 were prioritised. In the prioritisation model, the four main clinical data sources and their related features were used in prioritisation. The patients were prioritised on the basis of measurement outcomes from the integrated MLAHP and TOPSIS model.

3.2.1 Measurement results of the MLAHP model

To convert judgements, the MLAHP model performs mathematical calculations of the weights assigned by six experts to the four sources and their related features. The results of MLAHP measurements for the weight preferences of the six experts are listed in Table 3 (Kalid et al. 2018a, b). The overall consistency ratio (CR) was acceptable to all experts.

Table 3 Measurements of weights calculated for the six experts
Full size table

The weight preferences for each criterion and sub-criterion of the six experts were clearly different. Accordingly, the arithmetic mean for the final weights of the six experts was used to correctly eliminate variations among the weights they assigned to each criterion and sub-criterion. The arithmetic mean was then used in prioritising the patients by the TOPSIS model. The arithmetic means for the six experts are listed in Table 4.

Table 4 Final AHP weight for the arithmetic mean of six experts for each criterion
Full size table

3.2.2 Measurement results of the TOPSIS model

In the second stage of prioritisation, the TOPSIS model was used to prioritise the available patients at each level. The final weights of each criterion and sub-criterion are presented in Table 4. The final weights were used in prioritising the patients. Each patient was ranked according to these weights. The prioritisation results for all patients within each level are summarised in Tables 5, 6, 7 and 8 for patients who are at high risk (n = 66), require urgent care (n = 151), sick (n = 260) and in cold state (n = 23 patients), respectively.

Table 5 Results of prioritising high-risk patients
Full size table
Table 6 Results of prioritising patients who require urgent care
Full size table
Table 7 Results of prioritising sick patients
Full size table
Table 8 Results of prioritising patients in a cold state
Full size table

Each table presents the sequence of each patient in the dataset, the classification level for each patient, the prioritisation score according to the TOPSIS model and their order from the highest to the lowest values. A high value indicates that the patient should be prioritised.

4 Validation and evaluation

The proposed framework was validated and evaluated in this section. The validation process is presented in Sect. 4.1. The optimisation results were objectively validated on the basis of different features. The process by which the proposed framework was evaluated based on scenarios and checklist benchmarking is illustrated in Sect. 4.2.

4.1 Validation

The results were objectively validated. Data are expressed as mean ± standard deviation to ensure systematic ranking of patient prioritisation, which are calculated using Eqs. (10) and (11):

$$\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}=\frac{1}{n}\sum _{i=1}^{n}{x}_{i}.$$
(10)
$$\mathrm{S}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{d}\mathrm{a}\mathrm{r}\mathrm{d} \; \mathrm{D}\mathrm{e}\mathrm{v}\mathrm{i}\mathrm{a}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}=\sqrt{\frac{1}{N-1}\sum _{i=1}^{N}{\left({x}_{i}-\stackrel{-}{x}\right)}^{2}}$$
(11)

The prioritised patients were divided into three groups for each classification level to validate the prioritisation results. The patient’s number within each group varies from one group to another, and the number of groups or patients within each group does not affect the validation results (Albahri et al. 2020a, b; Almahdi et al. 2019; Alsalem et al. 2019; Mohammed et al. 2020a). Validation was conducted using a statistical platform based on the two methods (Salih et al. 2021). The mean ± standard deviation for each data row was measured for each group after normalisation and weighing, and the first group obtained the highest score. Assuming that the first group has the highest mean ± standard deviation, a comparison with the two other groups is considered to validate the result. The mean ± standard deviation of the second group must be lower than or equal to that of the first group. Lastly, the mean ± standard deviation of the third group must be lower than that of the first and second groups or equal to that of the second group. On the basis of the results of systematic ranking, the first group should be statistically proved to be the highest group among all groups. The results of the statistical analyses of the three groups of prioritised patients within each triage level are summarised in Table 9.

Table 9 Statistical analysis results
Full size table

The three groups were compared, and results revealed that the first group was the best and highest group, followed by the second group. Thus, the proposed framework that prioritised patients within each triage level tho underwent systematic ranking was valid. In conclusion, the prioritisation of patients in need of urgent care underwent systematic ranking.

4.2 Evaluation

Patients outside ED have been classified and prioritised to target other types of patients. Providing scenarios which represent all the environments and circumstances which may arise during patient prioritisation is important. Each scenario reflects certain issues which must be identified and addressed in classification and prioritisation. These issues were considered as comparison points for the proposed framework with the most relevant existing classification and prioritisation frameworks in checklist benchmarking. Comparisons were made on the basis of whether or not the comparative methods addressed the issues for each scenario as in previous studies (Abdulkareem et al. 2020; Ibrahim et al. 2019). Evaluation is illustrated in this section to determine the performance of the proposed framework. The proposed framework was compared with the most relevant frameworks in this area according to specific issues. Two scenarios are presented to determine the comparison points and issues in checklist benchmarking.

In scenario 1 (classification scenario), numerous patients are assumed, which is expected during disasters and accidents. The patients are classified according to emergency or classification level. Patients who may be very sick (high-risk patients) are assumed to be immediately attended to, but they may probably die even with intensive care. As a result, the other less sick patients (i.e. in urgent need to medical attention, sick, in a cold state and normal) will not receive immediate care, which may exacerbate their condition and lead to their death. Triage results in the best outcome for the greatest number of people when classification is performed properly. After the patients are classified, they are prioritised as urgent in each classification level. The record of each patient must be sent immediately to the hospital server for remote service and treatment. Therefore, the scalability issue and the ability to handle large amounts of data must be supported.

The second scenario (i.e. prioritisation scenario) has three sub-scenarios. In the first sub-scenario, numerous patients are present because of several situations, namely, population ageing, disasters and MCIs, in a specific area. In one perspective, this area has a large population of elderly patients who are remotely monitored by their providers. In another perspective, these remote patients are critically affected by MCIs and disasters (Wyte-Lake et al. 2016). The server of a hospital or a healthcare agency which monitors these patients must assess the situation and prioritise them according to the urgency of their medical condition. Therefore, prioritisation should support multi-criteria ranking considering the scalability issue and the ability to handle large amounts of data. Furthermore, the targeted tier and the environment where prioritisation is executed must be identified to prioritise patients from the most to the least urgent case to provide appropriate services and treatments. In the second sub-scenario, two or more patients at home must be prioritised with a slight difference in their healthcare emergency conditions. In this case, healthcare providers face the problem of recognising slight differences between the patients in terms of certain vital signs, regardless of the time needed to prioritise their patients. Traditional classification and prioritisation methods allocate such patients within the same scale or category (Claudio et al. 2014). Therefore, in prioritisation, the smallest difference between two patient records should be considered, and the accuracy of prioritisation must be improved. Thus, the order by which patients are attended to must be provided via supporting feature-weighting methods and multi-criteria rankings. In the third sub-scenario, two patients have different emergency conditions and send their requests to the server side at different times. The patient who has a less urgent situation sends the request first. The urgent situation should be prioritised first over the other less urgent situations. In addition, FCFS cannot serve patients in such a situation and may jeopardise the lives of patients. Moreover, FCFS cannot be used in reality (Claudio and Okudan 2010; Tan 2013). Therefore, methods for prioritising patients should consider all the conditions of all patients via feature-weighting methods.

The scenarios and related issues identified herein are regarded as comparison points in checklist benchmarking. The descriptions of each checklist comparison point are as follows:

  • Support vital signs: This point emphasises that vital signs are used in classification and prioritisation. Vital signs are important in evaluating a patient’s condition (Sakanushi et al. 2013; Salman et al. 2014).

  • Support chief complaint: This point considers the main complaints and uses them in patient classification and prioritisation because remote healthcare monitoring requires non-sensory data (Salman et al. 2014).

  • Targeted tier: This point identifies the level of classification and prioritisation. Three levels of architecture are available for remote healthcare monitoring and telemedicine, namely, sensors (Tier 1), base station (Tier 2) and remote server (Tier 3) (Claudio et al. 2014; Salman et al. 2014).

  • Remote environment: This point indicates whether or not classification and prioritisation are performed in a remote environment. Prioritisation is important for the continuous care of remote patients in a pervasive environment (Sarkar and Sinha 2014). The overwhelming heterogeneity of patient data in remote environments causes problems in classification and prioritisation.

  • Prioritisation in terms of category/order: This type of prioritisation shows whether or not categories or order methods support prioritisation. The categorisation method classifies and prioritises patients according to prioritisation level. Simultaneously, the order method classifies patients according to their emergencies. Most triage systems classify patients as prioritises, and patient order is usually determined according to the FCFS principle (Claudio et al. 2014).

  • Feature weighting: This point shows the weighing technique used. A server which scores a patient can provide more weight to the vital features over other features of less interest. In addition, the experts’ judgements and preferences are important in extracting the weights of vital signs (Abbasgholizadeh Rahimi et al. 2015; Claudio and Okudan 2009).

  • Multi-criteria ranking: This point shows whether or not a study deals with multiple criteria during prioritisation. Patient prioritisation is a complex problem in decision making (Ashour and Okudan 2010; Claudio and Okudan 2009; Göransson et al. 2008; Seising and Tabacchi 2013), and the decision is made based on a set of attributes (Faulin et al. 2012).

  • Handling large amounts of data: This point involves the handling of large numbers of patients with overwhelming data from multiple sources. Supporting large amounts of data is important because an overwhelming amount of data can be used to decide which patient should receive care first difficult (Sarkar and Sinha 2014).

  • Patient prioritisation accuracy: This point represents the accuracy of patient prioritisation. In the exact ranking of patients, prioritisation accuracy is reflected in the urgency of the situation with regard to the medical condition of patients. Accuracy also reflects the recognition of slight differences among patients during an urgent situation.

After recognising and defining the comparative checklist issues, the classification and prioritisation framework proposed herein was compared with those from other relevant studies. Each scenario had a 50% value divided by issues identified from each scenario. In the first scenario, six issues were identified, and the score for each issue was 8.33. In the second scenario, 10 issues were identified, and the score for each issue was 5. The comparison of checklist issues between the proposed work and those of benchmark studies is presented in Table 10.

Table 10 Benchmarking checklist
Full size table

The scenarios covered were investigated by addressing the comparison points of each scenario to compare the proposed and benchmark frameworks. With regard to the first scenario, all issues had already been covered by the benchmark studies (Albahri et al. 2019a, b) and the proposed framework. The issues covered by Salman et al. (2014) include targeted tier, scalability, remote environment and patient classification. The issue of handling large amounts of data for classification was not covered by Salman et al. (2014). Moreover, all issues of this scenario were not covered by previous prioritisation benchmark studies (Kalid et al. 2018a, b; Salman et al. 2017). Thus, the score for the first scenario of the proposed framework and benchmark studies (Albahri et al. 2019a, b) is 50%. The score for the benchmark of Salman et al. (2014) is 41.67%. The scores of both benchmark studies of Kalid et al. (2018a, b) and Salman et al. (2017) are 0%.

With regard to the second scenario, all issues were covered by the proposed framework and earlier benchmark prioritisation frameworks (Kalid et al. 2018a, b; Salman et al. 2017). In these three approaches [i.e. the proposed work and that of (Kalid et al. 2018a, b; Salman et al. 2017)], the prioritisation process performed is in Tier 3 (server), and each patient is compared with the other patients in the server. The MLAHP technique is used to determine the importance of each source and feature relative to other sources and features. Afterwards, pairwise comparisons are performed on the experts’ judgements to assign a fixed weight for each feature. The judgements of six experts were considered in setting the weights. A decision matrix is used to accommodate large amounts of data by listing alternatives which represent the patients (in column) and multiple criteria which represent the features used to evaluate the patients. Both approaches also adopted the MCDM model to deal with multiple heterogeneous sources from patients. The patient evaluation process should recognise that these features have different effects on patient evaluation. Patients are prioritised via simultaneous consideration of multiple attributes (vital signs and complaints) with respect to the proper weight assigned for each attribute to score the patients according to the most urgent cases without considering the FCFS technique. Thus, the MCDM model ranks and provides the order by which patients would be attended to on the basis of the urgency of each case regardless of request time. However, the issues already covered by a previous classification benchmark (Salman et al. 2014) are support vital signs, support chief complaints, support scalability, remote environment and prioritisation in terms of category/order. The issues not yet covered are (a) targeted tier as the prioritisation process performed in Tier 2 (base station) and assigning a PC value without comparing it with that of other patients in the server; (b) feature weighting as the set and test technique is applied for five diseases according to the dataset of patients without specifying whether the overall or part of the data set is used in the set and test method (in addition, a dataset may not reflect all cases which show the exact weights for each feature); (c) multi-criteria ranking due to data fusion has been applied to estimate the current medical condition of patients from various sources (however, data fusion is particularly difficult if the input data are heterogeneous [non-commensurate]); (d) handling large amounts of data; and (e) accuracy of patient prioritisation because patients are prioritised using a PC ranging from 0 to 100. However, only patients within this range can be prioritised, whereas patients with the same PC in the server are sorted in descending order via the FCFS principle. However, this principle cannot be used in reality because the situation of some patients may be more urgent than those of other patients who arrived at the ED earlier. Moreover, the benchmark studies of (Albahri et al. 2019a, b) did not cover any of the aforementioned issues in prioritisation scenarios.

The score for the second scenario of the proposed framework and the benchmark prioritisation frameworks (Kalid et al. 2018a, b; Salman et al. 2017) is 50% because they already covered all issues in this scenario (i.e. prioritisation). The score of the classification benchmark of Salman et al. (2014) is 25%, and that of classification benchmark studies of Albahri et al. (2019a, b) is 0%.

On the basis of the results of both benchmarking scenarios (i.e. classification and prioritisation), the total score and performance of the proposed framework is 100%; by contrast, that of the benchmark classification study of Salman et al. (2014) is 66.67%, and that of the benchmark classification studies of Albahri et al. (2019a, b) is 50%. The difference between the performance of the proposed framework and the classification study of Salman et al. (2014) is 33.33%. The difference between the performance of the proposed framework and that of previous studies (Albahri et al. 2019a, b; Kalid et al. 2018a, b; Salman et al. 2017) is 50%.

5 Conclusion

This study developed an emergency classification and prioritisation framework for managing patients with CHD who engage in remote health monitoring systems. In the telemedicine architecture, the improvements were achieved by remotely classifying and prioritising patients with CHD within each category in the server side (Tier 3). The patients are categorised into different emergency levels, namely, high risk, in need of urgent care, sick, in a cold state and normal, by using Dempster–Shafer theory. The patients within each emergency level are prioritised using a hybridised model of MLAHP and TOPSIS. The MLAHP model is utilised to extract subjective weights from six medical experts for each medical source. The TOPSIS model is employed to prioritise the available patients within each emergency level according to weights extracted from the MLAHP model. The validation and evaluation of the proposed framework were then achieved properly. The proposed framework can be used to increase the performance of classification and prioritisation in telemedicine environments and improve healthcare management. Recommendations for future works are multi-faceted. CHD complications are determined by the type of diabetes the patient is suffering from. Future studies may consider one or both types of diabetes. Thus, different classification and prioritisation models for diabetes must be investigated.