The use of intelligent techniques in medicine has brought a ray of hope in terms of treating leukaemia patients. Personalized treatment uses patient’s genetic profile to select a mode of treatment. This process makes use of molecular technology and machine learning, to determine the most suitable approach to treating a leukaemia patient. Until now, no reviews have been published from a computational perspective concerning the development of personalized medicine intelligent techniques for leukaemia patients using molecular data analysis. This review studies the published empirical research on personalized medicine in leukaemia and synthesizes findings across studies related to intelligence techniques in leukaemia, with specific attention to particular categories of these studies to help identify opportunities for further research into personalized medicine support systems in chronic myeloid leukaemia. A systematic search was carried out to identify studies using intelligence techniques in leukaemia and to categorize these studies based on leukaemia type and also the task, data source, and purpose of the studies. Most studies used molecular data analysis for personalized medicine, but future advancement for leukaemia patients requires molecular models that use advanced machine-learning methods to automate decision-making in treatment management to deliver supportive medical information to the patient in clinical practice.
Molecular cytogenetics of hematological malignancies and therapies is under development. Leukaemia is a hematological disorder where two leukaemia patients who may appear identical morphologically may have different molecular profiles and thus the variation in response to the prescribed therapies would be unpredictable [
The most common modes of treatment for leukaemia involve chemotherapy, radiation therapy, stem cell transplantation, and immunotherapy with interferon [
Therefore, the use of intelligent techniques in medicine has brought a ray of hope in terms of treating leukaemia patients. Intelligent techniques are able to conduct automatic searches to discover knowledge and learn from data to facilitate the task and achieve the objective. The broad areas frequently defined under intelligence techniques are as follows: knowledge discovery, machine learning, and data mining. These areas use statistics and probability to detect patterns that are difficult to study manually. Intelligent techniques will integrate various molecular technologies and sources of data, information, or knowledge to facilitate the development of personalized medicine and decision-making by physicians.
The personalized decision support system requires personal information or genetic information, such as genetic tests and medical tests, for each patient to integrate, as far as possible, the knowledge gained from genomics research relating to the disease in question [
Personalized medicine support systems can use available knowledge resources to deliver just-in-time information to individualize therapy. The existing pharmacogenomics knowledge base (PharmGKB) (available at
Developing personalized medicine support systems in some medical applications has already made significant progress. First, in cardiovascular diseases, many factors could influence cardiovascular disease, such as genes, environment, and lifestyle (exercise and nutrition). It was important to develop models for prevention, treatment management, or detecting disease to assist clinicians in treating cardiovascular patients. Indeed, the personalized decision support system for cardiovascular patients was constructed using two models. One model was for risk assessment using patients’ personal information, and the other was for generating advice to clinicians based on the first model’s results [
One angle of personalized medicine is to identify the correct disease subtype and patient classification. Machine-learning techniques were proven to achieve a high performance classification in identifying patient subtypes by using a support vector machine (SVM) and uncertainty SVM [
Medical researchers continue to emphasise that their studies are updated with the most effective treatment protocols, which could be used to treat leukaemia patients. The current system for achieving personalized medicine in leukaemia has been established by using the predictive factors to determine upfront treatment. Many groups of researchers have conducted studies by using different techniques to investigate several factors that could affect the drug responses. Studying a single biomarker as a predictive substance could indicate the response pretreatment and predict the risk to the individual [
According to current knowledge, many leukaemia researchers have applied intelligent techniques, but no reviewers have yet undertaken a systematized literature review from a computational perspective concerning the development of personalized medicine intelligent techniques for leukaemia patients using molecular data analysis. This review studies the published empirical research on personalized medicine in leukaemia and synthesizes findings across studies related to intelligence techniques in leukaemia, with specific attention to particular categories of these studies to help identify opportunities for further research into personalized medicine support systems in one category of leukaemia, namely, chronic myeloid leukaemia. A systematic search was carried out to identify studies using intelligence techniques in leukaemia and to categorize these studies based on leukaemia type and also the task, data source, and purpose of the studies. Our review is conducted to support health informatics and biomedical and bioinformatics in order to answer specific technical questions to help develop future research into leukaemia from a technical perspective.
Ten databases were searched, including Scopus, PubMed, Web of Science, BIOSIS, Inspec, MEDLINE, Embase, Springer, ACM Digital Library, and IEEE Xplore. The review was restricted to English-language studies published from 2001 to October 2016 because, prior to 2001, molecular targeted therapies and molecular responses for personalized medicine were not approved by the FDA for medical treatments in leukaemia but became more popular around this time [
The resulting abstracts were evaluated for inclusion. Then, the full text of those identified as meeting the criteria were obtained. Studies were included in the review, if the study used molecular data from adult leukaemia patients; used intelligence techniques to achieve the purpose of the study; was implemented as a model for adult leukaemia patients; was published in a peer-reviewed journal between 2001 and 2016; was published in English.
Because the intention was to review the literature to identify whether opportunities currently under clinical development are related to model analysis molecular data for personalized medicine in leukaemia, articles were excluded, if they published decision-analytic models for economic purposes; used pediatric leukaemia data; studied a nonpatient population; were not written in English; were doctoral dissertations or pilot studies; did not include the full text of the study report.
In total,
Flow chart showing the article-selection process.
55 studies described 55 unique intelligent techniques (Table
Of the commonest leukaemia types (Figure
Summary of the frequency of studies based on leukaemia type.
Some studies [
Emphasis has been placed on CML as a research opportunity because of developments in monitoring CML patients’ molecular response to molecular targeted therapy. The Australian Institute of Health and Welfare (AIHW) classified myeloid cancers as the 9th most commonly diagnosed cancer in 2016, with around 3,624 cases in Australia [
In CML, according to White and Hughes [
Microarray technology is an area of considerable focus for the purpose of cancer diagnosis (Figure
Summary of the frequency of studies based on data sources.
A huge opportunity arises from integrating data sources such as image data, clinical data, lifestyle or family history, SNP, gene-expression profiles, proteomics profiles, and metabolomics profiles. For example, SNPs have been investigated in an attempt to determine the susceptibility rate of patients suffering from leukaemia, which can support cases where patients have been diagnosed with leukaemia. The use of SNPs made it possible for physicians to predict the likely survivability of their patients after treatment, which is useful in determining the most suitable medical interventions.
In terms of patient care and administration, electronic health records (EHR) are often reused in research to answer specific research questions [
Yu et al. [
The other important source that has not attracted much interest in leukaemia studies is the data resulting from clinical trials studying healthy populations or epidemiological studies. Future development of clinical decisions can be guided by lessons learned from previous trials. Late-phase clinical trials (phases II, III, and IV) are considered to be massive sources of information that can be used to build personalized models. There is also a rapid increase in the number of electronic medical research databases that provide an opportunity for researchers to reuse medial data to create mathematical models.
The NCI [
The issue with the clinical trial data that it may be biased in several aspects: sampling, referral, selection, method, and clinical spectrum biases. Clinical trials may use sampling methods, sample size, and inclusion and exclusion criteria. Another aspect is referral bias where patients are referred by specialists and the data will represent preselected patients who have high prevalence of disease. Selection bias is clear when the clinical trial data includes groups based on a variety of demographics. In the method aspect, the data may be collected using different measurements, which leads to varying precision and specifications. Finally, the clinical spectrum bias represented in patient records may show other medical problems apart from the disease [
Medical research studies have several purposes, including classification of cancer types or distinguishing healthy cells from unhealthy cells for the purpose of diagnosis, identifying markers to help in the management of treatment, and determining the prognosis of risk. Managing leukaemia patients has gained attention since a successful study by Alvey et al. [
Among the 55 studies (Figure
Summary of the frequency of studies based on the purpose of the studies.
A new development is to extract relationships between biomarkers and the outcome in leukaemia patients. Focusing on CML, a predictive factor is a patient characteristic used to predict response to treatment [
The current methods used to identify risk in CML.
Previous methods | ||||
---|---|---|---|---|
Study | Factors | Method | Target prediction | Data and results |
Sokal score, Sokal et al. [ |
Age, spleen size (cm), blast (%), and platelets (109/L) | Multivariate analysis of survival | Risk groups for chemotherapy | Six European and American sources ( |
|
||||
Hasford score, Hasford et al. [ |
Age, spleen size (cm), blasts (%), eosinophils (%), basophils (%), and platelets (109/L) | Multivariate analysis of survival | Risk groups for interferon alpha alone | 14 studies ( |
|
||||
EUropean Treatment Outcome Study (EUTOS) Score, Hasford et al. [ |
Basophils (%) and spleen size (cm) | Multivariate analysis of response | CCgR at 18 months to Imatinib | Five national study groups ( |
|
||||
EUTOS Long-Term survival (ELTS) score, Hoffmann et al. [ |
Age, spleen size (cm), blast (%), and platelets (109/L) | Multivariate analysis of response | Long-term survival | ( |
Review of the studies, data sources, their purpose, and machine-learning algorithms reported from 2001 to 2015.
Study | Year | Tasks | Data source | Leukaemia types involved in the study | Purpose | Methods | |
---|---|---|---|---|---|---|---|
1 | Cho [ |
2002 | Feature selection and classification | DNA microarray | AML, ALL | Classifying leukaemia types | Pearson’s and Spearman’s correlation coefficients, Euclidean distance, cosine coefficient, information gain, mutual information and signal-to-noise ratio being used for feature selection |
|
|||||||
2 | Inza et al. [ |
2002 | Feature selection and classification | DNA microarray | AML, ALL | Classifying cancer, select genes related to cancer | Feature subset selection, case-based, and nearest neighbor classifier |
|
|||||||
3 | Farag [ |
2003 | Feature selection and classification | Blood cells image | AML, ALL | Classifying leukaemia types | A three-layer backpropagation neural network |
|
|||||||
4 | Futschik et al. [ |
2003 | Knowledge discovery | Gene expression | AML, ALL | Classifying leukaemia types and select gene expression | Knowledge-based neural networks and evolving fuzzy neural networks and adaptive learning and rule extraction |
|
|||||||
5 | Cho and Won [ |
2003 | Feature selection, classification, and ensemble classifiers | DNA microarray | AML, ALL | Classifying leukaemia types and select genes related to cancer | Correlation coefficient, Euclidean distance, cosine coefficient, information gain, mutual information, a feed-forward multilayer perceptron, |
|
|||||||
6 | Marx et al. [ |
2003 | Feature selection and classification | DNA microarray | AML, ALL | Classifying leukaemia from nonleukaemia | Principal component analysis and clustering |
|
|||||||
7 | Marohnic et al. [ |
2004 | Feature selection and classification | DNA microarray | AML, ALL | Classifying leukaemia types | Mutual information and support vector machine |
|
|||||||
8 | McCarthy et al. [ |
2004 | Knowledge extraction, classification, feature selection, visualization | Proteomic mass spectroscopy data, and gene expression | Melanoma, leukaemia | Cancer detection, diagnosis, and management | Naïve Bayes, support vector machines, instance-based learning ( |
|
|||||||
9 | Rowland [ |
2004 | Classification | Gene expression | AML, ALL | Classifying leukaemia types | Genetic Programming |
|
|||||||
10 | Markiewicz et al. [ |
2005 | Feature selection and classification | Images of different blast cell | Myelogenous leukaemia | Classifying patients | Support vector machine |
|
|||||||
11 | Tung and Quek [ |
2005 | Classification | DNA microarrays | ALL | Classifying leukaemia types | A neural fuzzy system, NN, SVM and the |
|
|||||||
12 | Nguyen et al. [ |
2005 | Classification | DNA microarrays | AML, ALL | Classifying leukaemia types | Support vector machine (SVM) |
|
|||||||
13 | Plagianakos et al. [ |
2005 | Feature selection and classification | DNA microarrays | AML, ALL | Classifying leukaemia types | artificial neural networks |
|
|||||||
14 | Li and Yang [ |
2005 | Feature selection and classification | DNA microarrays | AML, ALL | Classifying leukaemia types | SVM, ridge regression and Rocchio, |
|
|||||||
15 | Jinlian et al. [ |
2005 | Knowledge extraction | DNA microarray | AML, ALL | Leukaemia gene association structure | Clusters |
|
|||||||
16 | Diaz et al. [ |
2006 | Feature selection and classification | DNA microarrays | Acute Promyelocytic Leukaemia | Classifying Acute Promyelocytic Leukaemia (APL) from the non-APL leukaemia | Discriminant fuzzy pattern |
|
|||||||
17 | Feng and Lipo [ |
2006 | Feature selection and classification | DNA microarrays | AML, ALL | Acute leukaemia types |
|
|
|||||||
18 | Nguyen and Ohn [ |
2006 | Feature selection and classification | DNA microarrays | AML, ALL | Classifying leukaemia types | Dynamic recursive feature elimination and random forest |
|
|||||||
19 | Shulin et al. [ |
2006 | Feature selection and classification | DNA microarrays | AML, ALL | Classifying leukaemia types | Independent component analysis and SVM |
|
|||||||
20 | Chen et al. [ |
2007 | Feature selection, rule extraction, and classification | DNA microarrays | AML, ALL | Classifying leukaemia types | A multiple kernel support vector machine |
|
|||||||
21 | Ujwal et al. [ |
2007 | Feature selection and classification | DNA microarray | ALL | Identifying functional cancer cell line classes, classifying leukaemia from nonleukaemia |
|
|
|||||||
22 | Perez et al. [ |
2008 | Classification | Gene expression | AML, ALL | Classify leukaemia types | Hybrid fuzzy-SVM |
|
|||||||
23 | Yoo and Gernaey [ |
2008 | Feature selection and classification | DNA microarrays data | ALL | Classifying ALL origin cell lines from non-ALL leukaemia origin cell lines | Discriminant partial least squares, principal component and Fisher’s linear discriminant analysis, |
|
|||||||
24 | Avogadri et al. [ |
2009 | Knowledge extraction | Gene expression | Myeloid leukaemia | Discovering significant clusters | Stability-based methods |
|
|||||||
25 | Eisele et al. [ |
2009 | Knowledge extraction | Gene expression | CLL | Prognostic markers | Multivariate model |
|
|||||||
26 | Chaiboonchoe et al. [ |
2009 | Classification | DNA microarrays data | ALL | Identification of differentially expressed genes | Self-organizing maps (neural networks), emergent self-organizing maps (extension of neural networks), the short-time series expression miner (STEM), and fuzzy clustering by local approximation of membership (FLAME) |
|
|||||||
27 | Oehler et al. [ |
2009 | Knowledge extraction | Gene expression | CML | Identifying molecular markers | Bayesian model averaging |
|
|||||||
28 | Corchado et al. [ |
2009 | Decision |
Exon arrays | ALL, AML, CLL, CML | Classifying patients who suffer from different forms of leukaemia at various stages | Principal components, clustering, CART |
|
|||||||
29 | Glez-Peña et al. [ |
2009 | Feature selection and classification | DNA microarray | AML | Classifying gene expression | Fuzzy pattern algorithm |
|
|||||||
30 | He and Hui [ |
2009 | Classification | DNA microarray | ALL, AML | Classifying leukaemia types | Ant-based clustering (Ant-C) and an ant-based association rule mining (Ant-ARM) algorithms |
|
|||||||
31 | Mukhopadhyay et al. [ |
2009 | Feature selection and classification | DNA microarray | ALL, AML | Classifying leukaemia types | GA-based fuzzy clustering, neural network, and support vector machine |
|
|||||||
32 | Torkaman et al. [ |
2009 | Classification | Human leukaemia tissue | ALL, AML | Determining different CD markers | Cooperative game |
|
|||||||
33 | Zheng et al. [ |
2009 | Feature selection | DNA microarray | ALL | Gene ranking | Knowledge-oriented gene selection |
|
|||||||
34 | Mehdi et al. [ |
2009 | Knowledge acquisition | Gene expression | ALL, AML | Pattern clustering |
|
|
|||||||
35 | Porzelius et al. [ |
2011 | Feature selection, classification | Microarray and clinical data | ALL | Risk prediction | Feature selection approach for support vector machines as well as a boosting approach for regression models |
|
|||||||
36 | Chen et al. [ |
2011 | Feature selection, data fusion, class prediction, decision rule extraction, associated rule extraction, and subclass discovery | DNA microarray | ALL, AML | Select gene, classify leukaemia types, rule extraction | Multiple kernel SVM |
|
|||||||
37 | Gonzalez et al. [ |
2011 | Classification | Bone marrow cells images | ALL, AML | Classifying leukaemia subtypes | Segmentation method to obtain leukaemia cells and extract from them descriptive characteristics (geometrical, texture, statistical) and eigenvalues |
|
|||||||
38 | Tong and Schierz [ |
2011 | Feature selection and classification | DNA microarray | ALL, AML | Classifying two-class oligonucleotide microarray data for acute leukaemia | Hybrid genetic algorithm-neural network |
|
|||||||
39 | Chauhan et al. [ |
2012 | Classification | Genotype | ALL, AML | Identifying gene-gene interaction | Classification and regression tree |
|
|||||||
40 | Escalante et al. [ |
2012 | Feature selection and classification | The morphological properties of bone marrow images | ALL, AML | Classifying leukaemia subtypes | Ensemble particle swarm model selection |
|
|||||||
41 | Yeung et al. [ |
2012 | Feature selection and classification | Gene expression | CML | select gene, and predicted functional relationships | Integrating gene expression data with expert knowledge and predicted functional relationships using iterative Bayesian model averaging |
|
|||||||
42 | Manninen et al. [ |
2013 | Classification | Flow cytometry data | AML | Prediction method for diagnosis of AML | Sparse logistic regression |
|
|||||||
43 | El-Nasser et al. [ |
2014 | Classification | DNA microarrays | ALL, AML | Classifying leukaemia types | Implement enhanced classification (ECA) algorithm, SMIG module, and ranking procedure. |
|
|||||||
44 | Singhal and Singh [ |
2015 | Feature selection and classification | Image based analysis of bone marrow samples | ALL | Classifying leukaemia subtypes | Multilayer perceptron (MLP), linear vector quantization (LVQ), |
|
|||||||
45 | Yao et al. [ |
2015 | Feature selection and classification | DNA microarrays | ALL, AML, the mixed-lineage leukaemia (MLL) data | Classifying leukaemia subtypes | Random forests and ranking features |
|
|||||||
46 | Rawat et al. [ |
2015 | Computer-aided diagnostic system, feature selection, and classification | Bone marrow cells in microscopic images | ALL | Diagnosis lymphoblast cells from healthy lymphocytes | Support vector machine |
|
|||||||
47 | Kar et al. [ |
2015 | Feature selection and classification | DNA microarrays | ALL, AML, the mixed-lineage leukaemia (MLL) data | Classifying leukaemia subtypes | Particle swarm optimization (PSO) method along with adaptive |
|
|||||||
48 | Li et al. [ |
2016 | Classification | Gene expression | AML | Identifying feature genes | Support vector machine (SVM) and random forest (RF) |
|
|||||||
49 | Dwivedi et al. [ |
2016 | Classification | Microarray gene expression | ALL, AML | Classifying leukaemia subtypes | Artificial neural network (ANN) |
|
|||||||
50 | Krappe et al. [ |
2016 | Classification | Image based analysis of bone marrow samples | Leukaemia | Diagnosis of leukaemia and classifying 16 different classes for bone marrow | Knowledge-based hierarchical tree classifier |
|
|||||||
51 | Li et al. [ |
2016 | Classification | DNA microarrays | AML, ALL | Classifying leukaemia subtypes | A weighted doubly regularized support vector machine |
|
|||||||
52 | Ocampo-Vega et al. [ |
2016 | Feature selection and classification | DNA microarrays | AML, ALL | Classifying leukaemia subtypes | Principal component analysis and logistic regression |
|
|||||||
53 | Rajwa et al. [ |
2016 | Classification | Flow cytometry data | AML | Determining progression of the disease | Nonparametric Bayesian framework |
|
|||||||
54 | Ni et al. [ |
2016 | Classification | Flow cytometry data | AML | Analyzing minimal residual disease | Support vector machines (SVM) |
|
|||||||
55 | Savvopoulos et al. [ |
2016 | Knowledge extraction | CLL cells in peripheral blood | CLL | Capturing disease pathophysiology across patient types | Temporally and spatially distributed model |
Many studies [
Previously, two of the predictive factors closely involved in predicting the molecular response in CML were identified. The first such factor is IC50. In 2005, White et al. [
In practice, clinicians aim to treat individual CML patients with the most beneficial therapy. This can be made possible by using accurate risk assessment methods at diagnosis. When there is any doubt about either the diagnosis or the recommended treatment, a second opinion is often sought before considering any treatment. The need for multiple prognostic scores can occur frequently in a complex problem that has multiple independent experts with varying expertise. When developing prognostic scores have different patient populations, each score can capture different knowledge. There are two general major objectives for combining prognostic scores: first, one prognostic score enhances the decision of another one; and second, it increases the reliability of the final decision. However, integrating multiple prognostic scores could generate conflict in decisions and may not be sufficient to make a final decision.
It is important that clinicians are comfortable with a wide range of prognostic scores that will help to identify risk category because a conflict between scores may be observed in some patients. Consistency is defined as a score that does not contradict other prognostic scores. Consistency among prognostic scores can increase clinicians’ trust, as they rely on such results to make appropriate treatment decisions. It is important to study and understand the consistency of scores to help clinicians categorize patients into suitable risk groups and subsequently make better therapeutic decisions.
In light of the aforementioned aspects, it is necessary to conduct a study that can contribute to the CML medical field by solving the previous issues. Using machine-learning techniques and fusion techniques to address these problems could produce promising results. The first proposed solution is to build a personalized medicine support system as a predictive model to combine strong molecular, clinical data, and predictive assays for CML patients that could probably predict an individual molecular response. Moreover, predicting an individual response leads to predicting warning groups for each TKI. From a computer-science perspective, the above issues could be resolved by using a machine-learning algorithm that combines the most effective predictive indicators to predict the outcomes for each TKI, based on existing clinical profiles for individual CML patient characteristics. The main goal of this review is to improve the ability to manage CML disease in individual CML patients. Therefore, CML is an example of a research opportunity to predict the molecular response to TKI treatment. Using intelligent computing techniques could bring about promising results for CML patients.
Most of the studies that used machine learning and data mining incorporated two major tasks: feature selection and classification (Figure
Summary of the frequency of studies based on the task.
Many studies [
Knowledge extraction or acquisition has been a great challenge for researchers, as they exhibit unusual characteristics in many different genes relative to the number of tumor samples. AML acquires a similar appearance to ALL, which makes it nearly impossible for researchers to distinguish between synonymous patterns. However, Cho et al. [
Using multiple algorithms for knowledge extraction and classification has not attracted much interest from leukaemia researchers in previous studies [
Among the 55 studies, three groups of researchers [
From the review of studies based on the task, the need for personalized medicine in CML results in multiple active TKI therapies as molecular targeted therapy available for CML, multiple strategies utilized for frontline CML therapy, heterogeneity in responses, and multiple prognostic scores and predictive assays.
Therapy takes the form of two major strategies: (i) frontline Imatinib or (ii) frontline second-generation TKIs such as Nilotinib or Dasatinib [
Hematologic, cytogenetic, and molecular strategies for monitoring patient responses to therapies are used by European LeukaemiaNet [
Prognostic scores are used to personalize CML patient care by predicting responses to therapy. Although the prognostic scores (Sokal, Hasford, EUTOS, and the ELTS scores) remain in use today, they were developed either for identifying risk groups or for predicting cytogenetic response to therapy, but not for molecular response. Although two predictive assays,
Modern oncology is experiencing a paradigm shift toward personalized medicine, which aims to direct medical agents toward the tumor site. The field of molecular medicine is also undergoing transformational changes that are bringing a much needed revolution in healthcare. This breakthrough was made possible by technologies in genetic studies that led to the sequencing of the human genome. An analysis of biological samples from whole organisms has now been made possible. In addition, this invention has given a new lease of life to the treatment of cancer. However, the majority of cancer patients have been shown to develop adverse drug reactions due to overreliance on certain medications.
Intelligent techniques may be useful for clinicians in decision-making, warning of specific problems or providing treatment recommendations [
The use of personalized medicine support systems in medicine will bring a ray of hope to the treatment of leukaemia. Other frontiers of personalized medicine research, such as the role of genetics in infectious diseases, proteomics, epigenetics, and metabolomics, were not covered by this review and are out of scope of this research. This review was conducted based on current developments of personalized medicine support systems, and a systematized literature review was carried out on intelligent techniques using molecular data analysis in leukaemia. Both sets of literature led to identifying opportunities for further research for personalized medicine support systems in one category of leukaemia, namely, chronic myeloid leukaemia. We speculate that this paper will assist health informatics and biomedical and bioinformatics in order to answer specific technical questions to help develop future research into leukaemia from a technical perspective.
The funding agreement ensured the authors’ independence in publishing the report.
The authors declare that they have no conflicts of interest.
Haneen Banjar designed and performed the research, analyzed data, and wrote the manuscript. All the listed authors contributed substantially to drafts and revisions to the manuscript and approved the current revised version.
Financial support for this study was provided in part by a grant from King Abdulaziz University.