Bridging the Gap: Assessing the Integration of Artificial Intelligence in Healthcare for Improved Efficiency and Doctor Adaptability.

doi:10.21203/rs.3.rs-3592915/v1

Download PDF

Research Article

Bridging the Gap: Assessing the Integration of Artificial Intelligence in Healthcare for Improved Efficiency and Doctor Adaptability.

https://doi.org/10.21203/rs.3.rs-3592915/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

The premise behind implementing artificial intelligence in healthcare comes from the initiative that aims towards improving the healthcare standards, services, and accessibilities in a time and cost-efficient manner. The conceptual framework is based on the null hypothesis which states that there is no relationship between implementing artificial intelligence in healthcare and doctor’s redundancy. Having used the 2-tailed One-Sample T-Test to test that hypothesis on IBM SPSS, in order to successfully reject the hypothesis, and present an alternatively more accurate hypothesis that is more accurately representative. Additionally, DICE framework has also been presented in order to calculate the success/failure likelihoods of implementing artificial intelligence in healthcare. We conclude that artificial intelligence significantly impacts cost reduction in the long term, as well as only ever displacing behind-the-curve doctors who are unwilling to broaden their knowledge and develop their skills. Finally, we conclude that people’s resistance to change significantly affects the efficient and effective implementation of artificial intelligence in healthcare.

Management

Leadership and Ethics

Other Business

Health Economics & Outcomes Research

Alternative Hypothesis

Artificial Intelligence

Conceptual Framework

DICE Framework

Null-Hypothesis.

Semi-structured interviews were conducted in order to collect the data and responses, and consequently analyse to test the hypothesis in question. The interviews included 10 participants of which were all hospitals CEOs, giving the fact that they are the people with the power and authority not only to initiate the change, but plan and organise it as well. Interviews can be formed in several ways, including video conferences, phone calls or in person face to face interview. Interviews are crucial and significant tools in terms of collecting data, as they allow the researcher to detect an in-depth detail in addition to observe the body language of the participants. There are several advantages when conducting an interview, including accurate screening, meaning that the interviewer can instantly detect if the interviewee is providing false details about their race/gender/age. Another advantage when conducting a video or in person face-to-face interviews is that the interviewer could detect, capture, and analyse the interviewee’s emotions, change of tones, even change in vocabulary in order to understand deeper how the interviewee feels about a certain subject. The interviewer could also take advantage of conducting interviews by following up with supporting questions in order to gain more details and additional information from the interviewer. However, with advantages, always come disadvantages, in the case of conducting interviews, the process is generally on the high-cost side of the spectrum, it could also be time consuming for both the interviewer and the interviewee, in terms of the preparation, agreeing on the settings, time and place of the interview. One crucial disadvantage, that could be in certain circumstances turned into an advantage, is that the quality of the data collected during the interview, is relevant to the interviewer’s skills to collect the data well.

Sampling strategy:

It has been elected by the researcher that it is best to adopt the Purposive Sampling strategy for the interviews as part of the Non-Probability Sampling Methods, the reason being is that it is the most suitable method within the qualitative approaches when the researcher is intending to gain detailed knowledge on a certain phenomenon. Based on the researchers’ judgment and allows for easier data collection. This sampling method is often more popular with qualitative based research. It provides more in-depth data as it is more important in qualitative research as opposed to the generalisability in quantitative based research. Non-probability sampling method includes four different sampling strategies, as follows:

1- Convenience sampling strategy: this strategy generally includes whomever most accessible from the entire population. Time effective and the easiest strategy to carry out, however, the drawn sample is usually not representative, therefore it is not possible to generalise the results.

2- Voluntary response sampling strategy: not too different from the convenience strategy, in the sense that it is also based on the accessibility to the population. However, in this strategy, people willingly volunteer to partake in the study.

3- Purposive sampling strategy: the base of this particular strategy is the researcher’s judgment and expertise to choose the optimum sample that serves the purpose and objectives of the research. This strategy is usually used in qualitative based research when the researcher is intending to gain detailed knowledge on a certain phenomenon.

4- Snowball sampling strategy: this strategy comes to life when there is no ease of access to the population. As the name suggests, it only takes one participant to partake, then we can start recruiting participants through other participants, hence the snowball effect.

The interviewees, giving their extensive experience in management, were presented with the following question: Upon implementing artificial intelligence in healthcare, how much do you believe that would impact doctor’s redundancy?

Interestingly, in (Table I) that demonstrates distribution of the interviewees’ responses, none of the interviewees believed that artificial intelligence would have no impact, 2 participants believed the impact would only be marginal, 1 participant decided on moderate impact, 2 participants thought the impact would be notable, and %50 of the sample with 5 participants firmly believed the impact would be substantial.

Analysis / Hypothesis Testing.

One-Sample T-Test:

When using IBM SPSS, you are presented with several ways in which you can run a hypothesis test. However, not all hypothesis tests are valid for all types of data. The reason being is that certain hypothesis tests required certain type of data, certain number of groups and so on. Therefore, since an agreement was set between the interviewer and the interviewees that was based on the anonymity of the participants, on which the participants requested a total anonymity, which the interviewer guaranteed in lights of protecting the participants and their privacy, the only appropriate hypothesis test that we can run on the sample that we have is the 2-tailed One-Sample T-Test. The participants were presented with a scale alongside the questions they were asked, in order to be able to quantify their answers for the purposes of testing the hypothesis. The values of the scale ascended from 1–5, thus, since we are testing null hypotheses, the value of 1 on the scale which is represented with (none) on the given scale is going to be the test value, therefore, we will measure weather the responses from the 10 participants, regarding the certain questions which are intended to test the hypotheses, differ significantly from 1 which is the none response on the scale, which also sets up the test value at 1. Thus, µ = 1, and the conventional α = .05.

-Null Hypothesis:

H ₀₂ = There is no relation between implementing artificial intelligence and doctors’ redundancy.

-Alternative hypothesis:

H _A2 = There is a relation between artificial intelligence implementation and the redundancy of the behind-the-curve doctors.

The one-sample statistics table (Table II) shows the number of participants, represented with N = 10, and the mean with 4 which is significantly higher than none (1), which is the test value as mentioned before, which ultimately means the participants are leaning towards a significant impact of artificial intelligence on doctor’s redundancy. (Table II) also shows the standard deviation at 1.24722, and the standard error of the mean with .39441.

The one sample test table (Table III) is the main table of the analysis, as it shows not only the significance of running the test, but also the confidence interval. The t statistics under the null hypothesis, where there is no difference, meaning the mean is equal to a 1, we would expect t = 0. And so, (Table III) clearly shows that t = 7.606, and the question in a statistical sense, is that, does the t value lie far enough into the distribution, that we can confidently reject the null hypothesis? To answer that question, we need to interpret the p value. The degree of freedom, df, is 9, which is N-1. The p value, sig. (2-tailed), the reason behind conducting a 2-tailed test is because we need to detect deviations on either side of the null hypothesis. We are interested in rejecting the null hypothesis if we can obtain evidence of deviations above or below the given mean. The p = 001, therefore we have significant evidence to successfully reject the null hypothesis and accept the alternative hypothesis. Furthermore, what the significant value tells us is that the probability of observing the statistic value of t = 7.606 or more extreme under the null hypothesis, is .001, therefore we reject the null hypothesis. The last value we are going to interpret in (Table III) is the confidence interval, the mean difference is 3, which is the difference between the population mean of 1 and the sample mean of 4. There are two ways to interpret the confidence interval, and they are as follows: one way to interpret those values is in a test of the null hypothesis, as mentioned before, because we expected the mean difference to equal 1 under the null hypothesis, we can look at the confidence interval values and ask the question; where does 1 fall relative to the confidence interval values? Upon putting that question into experiment, 1 does not fall between the confidence interval values of 2.1078 and 3.8922. Therefore, that gives us another evidence to reject the null hypothesis at the .05 level of significance. The other way to interpret the confidence interval values is, in 95% of samples drawn from this population, this interval will capture the true population mean.

Using the APA significance reporting format: t (9) = 7.606, p = .001.

Artificial Intelligence in Modern Medicine:

Artificial intelligence is set to play a gradually significant part in the healthcare and medicine sector, the reason being is the result of the rapid development in terms of the availability of big data gathered from different medical records as well as medical wearable devices, in addition to the abilities of the supercomputers and the capability of learning algorithms. (Frost and Sullivan, 2016) claimed that at the time, the artificial intelligence healthcare market is rapidly growing at 40%, and it is anticipated to end up at $6.6 Billion in 2021. The abilities of supercomputers are also growing at a fast pace, part of the reason is the vast accessibility to (GPU) Graphic Processor Unit that facilitates and enhance the performance of the parallel processing mode. The cloud feature also supports and provides an unlimited storage for big data. According to (Bresnick, 2018A), it is crucial to learn algorithms since their accuracy and precision are constantly on the rise, and as a result of the interaction between those algorithms and the training data, they broaden the vision in terms of the range of treatments, patients results and outcomes as well as diagnostics. The constant developments and advancements of artificial intelligence in the age of big data can help physicians in terms of enhancing the efficiency, precision and accuracy of treatments and diagnosis. Artificial intelligence is capable of reducing costs by 50% and improving diagnosis accuracy by 30% (Hsieh, 2017). (Pratt, 2018) stated that there are many elements collaborated in order to support the developments and advancements of artificial intelligence in healthcare. Those elements consist of the abilities of supercomputers to process and analyse enormous sets of data, the large number of relevant collected data in the past few years, the growing spread of (EMR) Electronic Medical Records as well as the progress in terms of computing technologies altogether, which resulted in powering artificial intelligence developments in healthcare (Pratt, 2018). Although artificial intelligence in healthcare is still in its growing phases, its positive impact in healthcare is inevitable. For instance, (Pratt, 2018) observed that artificial intelligence has the ability to analyse the data collected from patients, wearable devices and home monitors and allow physicians to track the patients’ conditions in a way that the resources and time required would not be possible to achieve without the assist from artificial intelligence. Moreover, in accordance with (Bresnick, 2016), many experts have predicted that artificial intelligence will have a considerable effect in several departments in the healthcare sector including: decision making and underlying chronic diseases. (Hsieh, 2017) suggested that artificial intelligence is manifesting serious promises in pathology, cardiology, ophthalmology, and radiology.

Artificial Intelligence in Diagnostic Radiology:

The potential the artificial intelligence has in terms of analysing vast sets of data and derive helpful information proves its ability to revolutionise the process. (Kent, 2018) addressed the issue in terms of comparing the performance between the human being and artificial intelligence when he stated the fact that MRI machines, X-rays and CT scanners usually have an enormous number of complicated sets of data that consume significant amount of time for the human being to process and analyse. (Liew, 2018) believed that the implementation of artificial intelligence in the daily tasks in the healthcare sector is set to restructure the radiology practices. Artificial intelligence has the ability to assist the radiologists with clinical decisions in order to enhance the patients care quality. As far as image interpretation goes, the algorithms used in deep learning can help in terms of identifying, classifying, and categorising the patterns of diseases from the interpreted images in order to assist the radiologists with the best healthcare plan possible for the patient. As comparisons go, (Kooi et al, 2017) conducted an experiment in the mammography department between a standard computer aided detection system and a deep learning convolutional neural network algorithms that assist in detecting the lesions of breast cancer. Around 45000 images were used in order to train both systems, the convolutional neural network surpassed the standard computer aided detection system (Kooi et al, 2017). The results showed that the deep learning convolutional neural network algorithms performed at a radiologist level. (Liew, 2018) observed that artificial intelligence has the ability of segmenting images, detecting lesions and comparing patients’ history records. Although it is crucial to note that the practices and outcomes of the algorithms of deep learning are entirely dependent on the quality of the training sets that trained the system, this fact was also supported by (Lugo-Fagundo et al, 2018) when the authors stated that the quality and the quantity of those training sets are vital in regard to the advancements of the algorithms of the deep learning in radiology. With respect to the CT scans, the process necessitates a great deal of attention to thousands of scanned images. It was suggested by (Lugo-Fagundo et al, 2018) that the leading institutions in artificial intelligence should construct and start developing a standard sets of processes and guidelines that contain training sets of data. This type of cooperation is beneficial in terms of growing the amount of the sets of data available in order to train the algorithms of deep learning systems.

Artificial Intelligence in Pathology:

The pathology department as (Dyche, 2018) observed it merely relies on the pathologists trained eye in order to analyse a biospecimen slide and come up with the right diagnosis. As artificial intelligence keeps on developing and advancing, this task will eventually become impossible for the pathologists as a result of the variety of the diseases, their subtypes and their origins that appear in different genomics and biomarkers data. Furthermore, (Dyche, 2018) affirmed that on average, fewer medical students are choosing to follow a pathology career. Having said that, the approaches that are based on the algorithms of deep learning have a remarkable part to play. For instance, an experiment was undertaken by researchers at Google, the purpose of the experiment was to train a deep learning algorithms that were based on convolutional neural network in order to help in detecting a lymph node tissue with a mass of metastatic breast cancer from a biospecimen slide with a level of accuracy that matches the one accomplished by pathologists. Pathology is a highly demanding job, for human pathologist in particular as the process of detecting tiny masses of cancer on biospecimen slide can be daunting for them. On the other hand, artificial intelligence systems do not experience any sort of tiredness or fatigue, so they do not have to compensate for any accuracy lost due to tiredness and fatigue as a result of scanning countless amount of biospecimen slides. Without the help from machine learning, (Dyche, 2018) claimed that given the anticipated amount of the big data emerging into the healthcare sector, there will come a point where pathologists will eventually find it almost impossible to keep track with the appearance of the modern biomarkers. Nevertheless, machine learning approach was tested on a patients’ kidney in order to correctly determine and foresee for how long the patient’s kidney will stay functioning considering the patient is suffering from a chronic kidney damage. In addition, a team of researchers from Boston university in the United States trained a deep learning algorithms that were based on convolutional neural network using a renal biopsies to foresee the functioning of a kidney. As a result, the researchers realized that the convolutional neural network algorithms had more accuracy and precision than the old-fashioned scoring system the pathologists use in the process of measuring the failure of a kidney. Therefore, (Bresnick, 2018B) determined that artificial intelligence systems are capable to assist in terms of the clinical decisions in all departments.

Artificial Intelligence in Ophthalmology:

Ophthalmology is perhaps the most advanced department in the healthcare sector in terms of implementing various modern approaches into diagnostic and surgeries. Those approaches consist of machine learning and deep learning that have the ability to revolutionize the department of ophthalmology. A system based on deep learning was developed by (Schlegl et al., 2018) that helps in detecting and quantifying automatically (IRC) intraretinal cystoid fluid and (SRF) subretinal fluid. According to (Schlegl et al, 2018) the purpose of the system is to accurately identify the patterns of the intraretinal cystoid fluid for the patients with wet (AMD) age-related macular degeneration or a (RVO) retinal vein occlusion in order to be able to distinguish the subretinal fluid from the intraretinal fluid. (Schlegl et al, 2018) came to the realization that adopting the approach of deep learning in the analysis of retinal images perfectly resembles the meaning of “for the differentiation in the process of detecting different types of retinal fluid across the well-spread macular diseases and (OCT) Optical coherence tomography devices”. In addition to that, (Liu et al, 2017) observed that artificial intelligence-based systems and machines are in constant development to facilitate the process of diagnosing based on the analysis of the images of the slit-lamp examination. (Ruiz et al, 2017) affirmed that it is significant for ophthalmologists to learn how to use artificial intelligence-based systems in order to raise the chances of accurate diagnosis and proceeding with the surgeries.

Artificial Intelligence in Cardiology:

As stated by (Weng et al, 2017), a study that was undertaken in the United Kingdom suggested that machine learning has the ability to reduce the risks associated with the diagnosis of cardiovascular diseases by correlating a set of complicated interactions between the risk factor associated. For training purposes and in order to associate the patients’ medical history with the rates of heart attacks, the researchers presented 295000 patients medical record data to several machine learning algorithms, including: neural network, logistic regression, random forest and gradient boosting machines. (Weng et al, 2017) claimed that the purpose of these algorithms was to foresee who of the additional 82000 patients might get a heart attack based on the medical record of the patients. The neural network-based algorithms had the best performance out of the other four mentioned algorithms, predicting and anticipating 7.6% more cases as it outperformed and surpassed both the American college of cardiology and the American heart association approaches with just 1.6% (Hsieh, 2017). (Hsieh, 2017) observed that 355 patients out of the 83000 patients’ records associated with cardiovascular diseases could have saved their lives. (Weng et al, 2017) came to the realization that machine learning considerably enhance the precision of the prediction of cardiovascular risks, improving the number of patients distinguished who could take advantage of the preventative treatments and refraining from any needless treatments for others. (Dawes et al, 2017) conducted another study that involved developing a machine learning algorithm that allowed physicians to accurately predict the results of patients who suffer from pulmonary hypertension. Around 250 patients medical record data were in use for the mentioned study. According to (Dawes et al, 2017), the machine learning algorithms helped in enhancing the survival rate of the standard risk factors associated with the patients who were diagnosed with pulmonary hypertension.

Future of Artificial Intelligence in Healthcare:

Predicting the future of artificial intelligence in the healthcare sector is difficult, although we can easily confirm that artificial intelligence can play a significant partner role in healthcare. This theory is built around certain attributes of artificial intelligence systems. (Sappin, 2018) suggested that Watson, the supercomputer developed by IBM can skim through countless data, search millions of medical records as well as surpassing the capability of any physicians in terms of knowledge and broadness. As stated before, tiredness and fatigue have a significant impact in terms of the judgements of the physicians, this could cause patients their lives if at some point a physician forgets that a patient might be allergic or vulnerable to a specific type of drugs along the process, while an artificial intelligence system will not compensate for that as the systems do not suffer from such a thing as tiredness and fatigue. It is a well-known fact that perhaps the strongest element of an artificial intelligence system is its ability to collect countless amounts of data as well as analysing them and come up with various conclusions. This vital feature can assist in diseases predictions, which is one of the most challenging tasks the physicians face. Artificial intelligence has the ability to help physicians to evaluate the risks of cancer and the risks of heart attacks in comparison with strokes. All the developments and advancement in artificial intelligence and machine learning as well as deep learning algorithms are here to help enhancing the experience of the patients and facilitate the physicians’ tasks on a day-to-day basis. As a result, that will make the practices in the healthcare sector more precise, accurate and perhaps lowering down the costs as a result of an accurate early diagnosis, correct treatment plans, in addition to preventing any needless tests. This is a revolution for all the parties involved in the process, physicians, patients, and the healthcare sector entirely.

Transforming Medical Care Through Artificial Intelligence:

The integration between the artificial intelligence approaches and the human physicians could result in enhancing and improving the way the medical care is delivered to patients in a better way than what artificial intelligence and human physicians can do on their own. Take into consideration the situation of precision medicine, which the main purpose of it is to personalize the treatments to fit each and every individual patients’ traits. The results of this situation are promising in terms of transforming the way the medical care is delivered to patients by allowing physicians to determine the best number of dosages of medication needed as well as to identify which gene mutation causes certain types of cancers. Another significant factors that are set to help in transforming the delivery of the medical care are artificial intelligence and machine learning which can process, analyse, and conclude from a countless number of images, specimen slides and clinical data, therefore assisting in enhancing the efficiency of the physicians, accuracy of diagnosis and individualize treatments. Artificial intelligence systems are set not only to enhance the efficiency and accuracy of diagnostics, but to improve the other departments within the healthcare sector. (Pearl, 2018) considered several numbers of artificial intelligence systems that can positively affect different departments in healthcare, including: radiology (MRI machines, X-rays, and CT scanners), pathology (diagnosis by microscopy), ophthalmology (detecting and quantifying intraretinal cystoid fluid and subretinal fluid in addition to any cardiovascular diseases). (American Cancer Society, 2017) observed that 50% of the females undergoing a yearly mammogram examination over a period of 10 years will result in a false positive. The results of false positive in mammogram tests cause some serious concerns to both the radiologists and the patients, which more often than not calls for unneeded examinations like ultrasounds and Magnetic Resonance Imaging known as MRI to ensure that there is no cancer, which eventually results in consuming more time than needed as well as the high costs of running those additional unnecessary tests (Griffiths, 2016). Furthermore, (Kakar, 2017) believed that artificial intelligence can highly impact the robotic surgeries, where the robots increase the accuracy, reduce blood loss and pain. (Wottawa et al, 2016) claimed that robotic surgery decrease the chances of (STI) soft tissue injuries, the reason being is that robots minimize the force of gripping as a result of tactile feedback that allows the surgeon to proceed with the surgery in a remotely manner. According to (Singh, 2018), all the various wearable healthcare gadgets from insulin monitors, blood pressure monitors and all the fitness related devices will significantly help in the delivery of the medical healthcare department as they will be directly connected to a central artificial intelligence database and constantly feed data into it. Thus, this process should allow the patients to monitor their medical condition at home, resulting in enabling the physicians to come up with the best possible treatments (Singh, 2018).

The Impact of Artificial Intelligence on Physicians:

It has been observed for number of years that artificial intelligence will outperform physicians in terms of effectiveness when dealing with image recognition and predictive analytics, as physicians cannot process nearly half the amount of data as artificial intelligence in a timely manner. Thus, the high anxiety among physicians, and radiologists in particular that artificial intelligence will soon displace them. This was supported by (Pearson, 2017) who stated that the simplest forms of artificial intelligence have the ability to decipher the most complicated images and specimen slides at the same level of accuracy as the most experienced radiologists, and will eventually takeover. In contrast, (Pearson, 2017) continued to say, that artificial intelligence will never entirely displace radiologists, but rather collaborate to maximize the outcomes. The need to assess and measure the artificial intelligence impact on physicians is a significant yet challenging task, (Recht and Bryan, 2017) shared their views of the matter as they observed that artificial intelligence will eventually be the biggest part of the radiologist’s day to day tasks, facilitating their jobs and increasing the value, accuracy and efficiency of their tasks. With respect to the roles that artificial intelligence will undertake in the radiology department, (Recht and Bryan, 2017) listed some of the tasks for artificial intelligence, including: segmentation and quantification, in order to allow physicians to focus more on other value-added tasks, including: being there for the patients and merging clinical teams in order to augment the patients care. Taking into account physicians in general, it is sensible to predict that artificial intelligence is here to assist physicians and not replacing them. This was discussed by (krittanawong, 2018) when he stated that as effective and accurate and efficient artificial intelligence is, it still lacks several serious aspects when it comes to the patient’s bedside! Meaning that artificial intelligence cannot interact with patients on an emotional level, therefore it is not capable of showing any empathy towards the patients, cannot gain their trust and reassure them, which are significant aspects in terms of the patient-doctor relationship. However, experts are speculating that this is highly subjected to change at some point in the near future as communication skills of artificial intelligence are being developed by Google when they revealed a phone call of an artificial intelligence assistant during their Input/Output conference in 2018. Having said that, (Krittanawong, 2018) suggested that physicians are still required to be there even for the conventional physical tests, neurology in particular, where a a high level of patient-doctor interaction is still required to do the job in the optimal manner. All in all, despite the fact that artificial intelligence will reach a point where taking a real-time CT scans in addition to all the other physical tests is a daily routine task, physicians are still required to participate in the challenging and ambiguous medical cases. Artificial intelligence systems are mainly based on priorities in the machine learning and deep learning cases, although those algorithms might not perform as anticipated in some of the unusual cases including treatments side effect and resistance of treatment when there is no available example in advance to build the conclusion on. As a result, those listed cases support the fact that artificial intelligence is only here to augment the physicians’ skills and facilitate their jobs, as well as increasing the efficiency and accuracy of their tasks, and highly improbable to displace the conventional relationship between patients and physicians. It is predicted that all the medical records, imaging and genomic data are going to expand even more as individualized medical care evolves. As a result, for the near predictable future, medical care has a high potential to depend more and more on data, given the collaboration between medical care and artificial intelligence. Taking into account this significant current orientation in modern medical care, medical schools are reconstructing their curriculums to go along with the new advancements of artificial intelligence systems. (Dyche, 2018) observed that medical schools are adding new machine learning, deep learning, data management and an overall technology infrastructure courses. Artificial intelligence is capable of supporting the medicine requirements in the future by analyzing the enormous amount of data that the healthcare institutions and the patients themselves provide to the mentioned central artificial intelligence database. One of the most important roles of artificial intelligence is to takeover some of the conventional tasks for physicians and enable them to have more time spent with the patients. As it is predicted that artificial intelligence is improbable to entirely displace physicians, at least in the near future, it is vital for physicians to understand the basics and fundamentals of artificial intelligence, in addition to how artificial intelligence will assist the physicians in order to provide the best results possible for the patients.

-Dice Framework.

(BCG’s DICE tool, 2022) The framework was developed by Perry K. Kathleen C. and Alan J, in the mid-1990s, current and former partners at Boston Consulting Group, that was founded in 1963. As (Tahir, 2020) explained, DICE factors stand for: Duration, Integrity, Commitment and Efforts. Those factors differentiate between organisations and the approach they adopt in terms of combining these factors in order to excel in the change process, and that would always separate organisations that are destined to succeed, to those destined to fail. The change scale, as it is the case for most scales, has two extreme points at either end of it, as well as a middle ground. One of those extreme points represents a short-term project, managed by an experienced, skilful, tenacious team, supported by the management, ought to succeed. The other extreme point represents a long-term project, led by unexperienced, unmotivated team with no support from the top management, that is inevitable to fail. Organisations are smart, and it does not propose any challenge for them to identify and pinpoint changes at either extreme of the change scale, though most organisations and indeed initiatives are more likely to settle for the middle ground where the possibilities of either success or failure are hard to judge. In conclusion, organisations must run a comprehensive analysis of the DICE factors prior to launching and initiating the change, in order to determine whether the anticipated change is a “High or Die” for the organisation.

Duration: the first mistake organisations usually commit is to be solely concerned with how long it will take in order to implement the change completely, in addition to believing that the longer the implementation process takes, the less likely for the change to succeed, and there starts the domino effect, where the initial momentum dies out, opportunities fade away, objectives become burdens, main backers and supporters lose their motivation, and issues will accumulate to the point where the project will inevitably collapse. Nevertheless, contrary to the popular belief, studies confirm that a long-term continuously assessed change, has a higher likelihood of succeeding than a short-term change that is not frequently evaluated. Therefore, the duration between frequently assessing the change, is far more crucial than the change span. When considering the magnitude of the change in terms of adopting artificial intelligence in healthcare, such a drastic and revolutionary change, organisations must assess the implementation process every fortnight, given the fact that the likelihood of troubles arising rises significantly when the duration between the assessments oversteps a month mark, taking into account the complexity and magnitude of the change. Management should be responsible for such calls, whether the assessments needed to be planned more or less frequently based on the managements views as for how long can change run before running into certain obstacles. On the other hand, less complex, more familiar, and orthodox changes, can be reviewed on a monthly basis. There are different approaches in which the management could undertake in order to assess the change implementation process, however, the best way by far is to setup goals and milestones and measure their effect. Those milestones cannot be of a day-to-day regular activities, as that would fail to represent the true significance and impact of the milestones, but rather the milestones that have exceptional sense of achievement. The sole reason behind that is to showcase to the management and sponsors if applicable and there has been a significant progress since the last assessment was administered. If case a milestone does not appear to be reached within its timetable, the team in charge of the change is responsible for, firstly, understanding the events that led for this particular milestone not to be met, correct it, learn from it, and ensure that it does not reoccur. Moreover, the manner in which those milestones are addressed is equally as important as reaching them. For instance, “consult stakeholders” does not represent the milestone in attractive and rewarding manner but stating that “consulting the stakeholders completed” indicates that something has been achieved and progress has been made.
Integrity: performance integrity entails to what extent are organisations willing to trust and rely on staff and management to successfully execute the change. Ideally, we would want every team to be perfect, however no organisation has that calliper of staff across the board to guarantee that. Additionally, managements are more often than not unwilling to permit staff to join in, out of fear that the daily activities might falter. Though, as successful change is normally based on the team’s quality, organisations have to create a free space for the star staff in which they can join in the change efforts without compromising the daily activities. Recent studies confirm that in the organisations where change has been successfully implemented, staff took the extra step and walked the extra mile to guarantee that the daily activities had been completed. It does not suffice for managements to only informally ask staff how the change implementation unfolding, they have to identify the staff roles, accountability, and commitment. Managements must select a leader for the team, and most importantly, figure out the structure of the team. Smart managements are highly comprehensive in terms of selecting teams. As they address potentials and talents by asking for names from the human resource department that they believe would match up the potentials and criteria the smart managements have identified. Smart managements normally tend to personally conduct the interviews with people, so that they are able to pinpoint the knowledge, potentials and set of skills those people have. In addition to that, managements should always make the public the parameter in which the management will assess the performances of teams and address how that assessment suits the appraisal process of the organisation. At times, managements make the odd mistake of believing that when an individual is good at what they do, liked across the board, they are automatically team leaders. That does sound rather sensible, though it is not always the case, as those type of people are not necessarily effective in terms of leading the change. A CEO who has effectively and efficiently led radical transformations in the past decade listed a set of particular criteria to assess managements about their decisions when it comes to the stage of electing a change leader, those criteria are; problem solving, methodical, results driven, work out of sight without drawing attention to themselves, organisationally intelligent, take decisions and responsibility for their results and consequences.
Commitment: there are two groups of people that organisations need to boost their commitments in order to get underway; the first group includes those who have certain influential attributes on others within the organisation, that does not necessarily mean they should be of top or executive management. The other group would be of those who have to deal and be in direct contact with the daily processes and the new installed systems and work schedules. This particular group is highly influenced by managements and the first mentioned group, therefore, a visible commitment from the management would be the catalyst for the commitment of the daily workers. It is inevitable that if the staff cannot sense the management’s support and commitment to change, it is highly unlikely for the workers to change their attitudes towards the change. We can argue that the rule of thumb here would be once you frequently start mentioning and bringing up the initiative of change, and try to manifest it into existence, managements will more likely take it into account and setup the initial practicalities of the initiative. At times, managements are hesitant, and indeed resistant in terms of supporting initiatives. Which is reasonable, giving the fact that if not calculated, analysed, and measured comprehensively and accurately, changes could cost workers their jobs, and consequently their livelihoods. Nonetheless, if managements fail to build bridges with their workers and communicate the necessity for the change with them, and what are the possibilities and potentials the change would bring, that could very well jeopardise the entirety of the change initiative and its implementation. Occasionally, organisations overlook the impact managements and staff bring into the transformation process, whether the reason be ineffective or inconsistent communication, which ultimately leads to repelling workers who are directly impacted by the change. Moreover, it is not a surprise how regularly managements think of a certain idea as a good idea, while workers believe it is not, or a certain order or a work memo to be clear and straightforward, while workers misunderstand it. This normally occurs when management deliver critical messages in different fashions and manners. An example of that event happening is when managements for instance survey their workers to measure their level of commitment to the change, as the results come back, the score was significantly lower than expected, the reason being that when management tried to communicate with their workers in terms of layoffs as a possible consequence of the change, a member of the management team addressed the issue by saying “layoffs are not happening”, which completely eliminate the layoff possibility and entails that no matter the outcome, the management is not laying-off any worker, while another member of the management team on a different occasion addressed the issue by saying “we are not expecting layoffs to happen”, which opens up the door for that possibility to happen.
Efforts: when organisations initiate the transformation process, they occasionally tend to underestimate the fact that workers are already overwhelmed with their daily tasks and work schedule. Workers in all kinds of different businesses work on average 70 hours a week, taking that into account, if organisations impose more responsibilities and tasks to managers and workers and expect them to cope and deal with the change to their daily working schedule and the system they use and familiar with, the least the managements and workers would do is to resist that change as a natural response. Organisations must take into account and carefully measure how far can workers go beyond their current schedules in order to for them to be able to contribute efficiently and effectively to the change. Workers should not have their workload increase by more than 10%, anything higher than that, obstacles are bound to raise, and resistance will start taking place, workers attitude will shift into negativity, and conflicts will pile up. Therefore, organisations must consider whether the additional responsibilities they lay on their workers are not too demanding on top of the already intense workload that they have to deal with, and whether the workers as a result are expected to show resistance. Organisations, prior to launching the transformation process, have to come to a conclusion as to whether it is possible to lay off certain routine responsibilities of the workers who are expected to have a significant impact on the transformation process, in order to free up some of their time to contribute to a greater cause, organisations can achieve that by laying-off certain redundant and inessential responsibilities of the workers. additionally, organisations should never neglect the other departments affected by the change, and frequently assess and address the departments that could have a critical role in facilitating the change. Other measures in which organisations can spread the responsibilities, ease the pressure, and facilitate the transformation process is to recruit temporary short-term people, who are capable of taking over some of the inessential and routine responsibilities that have been laid-off of certain workers who are more involved in the change process, until the transformation process is accomplished. Organisations should have this planned prior to launching the transformation initiative, for the simple reason that laying-off duties and responsibilities is usually costly, and postponing certain tasks is time consuming. Accordingly, organisations must work twice as hard to find the right balance between the two mentioned predicaments, otherwise, the consequences might just outweigh the desired end results. The next step will be building the framework that should assist and support organisations in terms of assessing the change initiative and pinpoint the areas in which there is a chance of improvement which would ultimately result in increasing the success likelihoods. There are certain variables that impact each of DICE factors, and points scoring systems is put in place accordingly. Organisations then would assign points to each of DICE factors and calculate and measure the end results in order to reach a verdict on the project success likelihoods.

Dice Scoring System:

organisations can calculate the success likelihoods of the change by assigning scores to each of DICE factors. The score system goes from 1–4, 1 being the highest and 4 being the lowest, this is crucial to remember and not have it confused the other way round. Consequently, the best score which entails the highest success likelihoods using that scoring system would be the lowest, and vice versa, the higher the score, the less likely for the change to succeed. Following are the guidelines for the framework:

Duration / D: In the instance of change duration, we ask the significant question; how frequently is change assessed?

We would give a score of 1 point if assessments frequently take place every 2 months or less. We would give a score of 2 points if assessments frequently take place every 2 to 4 months. We would give a score of 3 points, if assessments frequently take place every 4 to 8 months, and finally we would give a score of 4 points if assessments take place every 8 months and over. In the case a change duration is anticipated to be less than 2 months, the same principal applies, but with weeks rather than months as assessing periods.

Integrity / I: In terms of performance integrity, we ask the simple question; is the leader assigned to the project qualified and competent to lead and motivate workers?

We would give a score of 1 point, if the leader is qualified and competent, brings the best out of the workers, and is highly respected and regarded by them. We would give a score of 4 points, if the leader lacks competence, workers are demotivated and don’t follow the leaders’ instructions, and finally we would give a score of either 2 or 3 points to anything in between those two extremes of 1 and 4.

Commitment (Managements) / C1: The management’s commitment to change is crucial because it sets the tone for the how the project pans out. Therefore, we must ask few questions in that regard, in order to determine the level of commitment the managements have, some of those questions are as follow; do managements communicate the importance of the change effectively with the workers? are the potentials and end results clear in their positive impact on the workers livelihoods? Have managements dedicated sufficient time, resources, and efforts to the change?

We would give a score of 1 point if managements convincingly managed to cover all the mentioned bases. We would give a score of either 2 or 3 points, if managements managed to meet the best part of those requirements, and finally we would give a score of 4 points, if managements fail to show motivation and support to the change.

Commitment (Workers) / C2: As for the workers, who are at the operating end of the change, we must assure ourselves that they are onboard with the change, and so, we have to evaluate the workers stance on the change happening by asking those following questions; are the workers motivated and eager for the change? Do the workers understand the reasoning behind the change, and that its rewarding end results outweigh the uncertainties and discomfort along the way? We would give a score of 1 point if the workers are showing enthusiasm towards the change. We would give a score of 2 points if the workers are willing to take on the change, and finally we would give a score of either 3 or 4 points if the workers are showing some signs of resistance towards the change.
Effort / E: As explained earlier, workers should not have their workload increase by more than 10%, anything higher than that, it would backfire, obstacles are bound to raise, and resistance will start taking place, workers attitude will shift into negativity, and conflicts will pile up. Therefore, it is crucial that we address the issue, first and foremost, and then ask the question; what is the increased percentage of work the workers have to take in order to significantly contribute into the change? Does that work pile up on an already overfilled schedules for the worker?

We would give a score of 1 point if the incremental overwork is 10% or less. We would give a score of 2 points if the incremental overwork is between 10%-20%. We would give a score of 3 points, if the incremental overwork is between 20%-40%, and finally we would give a score of 4 points if the incremental overwork is more than 40%.

DICE Scores Calculating Formula

D + { 2 × I } + { 2 × C 1 } + C 2 + E

As demonstrated before, the score system goes from 1–4, 1 being the highest and 4 being the lowest, this is crucial to remember and not have it confused the other way round. Consequently, the best score which entails the highest success likelihoods using that scoring system would be the lowest, i.e.,7, and vice versa, the higher the score, the less likely for the change to succeed, i.e., over 19. Therefore, the DICE scores calculating formula generates scores as low as 7, and as high as 28. If the score was 7–14, the success likelihoods of the change are at their highest. If the score was 14–17, the success likelihoods of the change are not as strong, as risks start to rise, though success likelihoods still overshadow failure. If the score was 17–19, the success likelihoods of the change are doubtful as the change at this stage would show significant risks, but there is still potential of succeeding. If the score was over 19, the change initiative should be called off, as none of the success criteria would exist at that stage, comprehensive revision to the change initiative is necessary to understand what led the DICE score to be as high.

It is probable to make the case for whether artificial intelligence would completely displace health workers, or perhaps go in the other direction by assisting and creating new opportunities. There is a significantly rising interest in this predicament, as to what the reality and implications of implementing artificial intelligence in healthcare, and where does that leave physicians and surgeons. There is a school of thought that argues that the impact would be slightly marginal, supported by the fact that every technological breakthrough has created a new economic opportunities in one way or another, as technological advancements always tend to leverage people’s productivity. A trajectory by the World Health Organisation estimates a shortage of 13 million health workers by 2035, caused by a shift in interest among people in general in healthcare. We have experienced a glimpse of that issue when Covid was at its peak, as it drastically short of healthcare workers we are. Taking that into account, perhaps artificial intelligence in healthcare could be the catalyst for that change, and be that attracting prospect of shifting people’s interests back again into healthcare careers. Another fascinating theory suggests that to a certain extent, artificial intelligence will indeed displace certain healthcare workers, who are not willing to improve and develop their skills constantly. The evidence being that technology always comes around to replace workers on the lower strata, and if a healthcare worker is unable of moving ahead of the curve, giving that they were appropriately and fully trained, those are always the ones at risk of being replaced, regardless of it being by artificial intelligence or a more qualified worker who is eager to acquire new skills and constantly expand their intellectual capacity and capability.

Participants' consents were obtained prior to any data collection, as participants read the consent statement below and agreed to partake in the study.

'I agree to take part in the research. I understand what my role will be in this research, and all of my questions have been answered to my satisfaction. I understand that I am free to withdraw from the research at any time, without giving a reason. I am free to ask any questions at any time before and during the study. I understand what information will be collected from me for the study. For the purposes of the Data Protection Act (2018), if this project requires me to produce any personal data, I have read and understood how it will be processed. I understand what will happen to the data collected from me for the research. I have been informed how my data will be processed, how long it will be kept and when it will be destroyed. I have been provided with a copy of this form.'

Competing Interests:

The author declares no competing interests regarding the research, authorship, and publication of the paper.

Data Availability:

The datasets generated during and/or analysed during the current study are available from the corresponding author or reasonable request.

Ethical Approval:

I am pleased to inform you that your ethics application has been approved by the School Research Ethics Panel of the Engineering and the Built Environment (SREP-EBE) under the terms of Anglia Ruskin University’s Research Ethics Policy (Dated 1 May 2019, Version 1.10).

Informed Consent:

I agree to take part in the research. I understand what my role will be in this research, and all of my questions have been answered to my satisfaction. I understand that I am free to withdraw from the research at any time, without giving a reason. I am free to ask any questions at any time before and during the study. I understand what information will be collected from me for the study. For the purposes of the Data Protection Act (2018), if this project requires me to produce any personal data, I have read and understood how it will be processed. I understand what will happen to the data collected from me for the research. I have been informed how my data will be processed, how long it will be kept and when it will be destroyed. I have been provided with a copy of this form.

Author Contributions:

The author confirms sole responsibility for the following: study conception and design, data collection, analysis and interpretation of results, and manuscript preparation.

American Cancer Society. (2017). Facts and figures. Available at: <https://www.cancer.org/content/dam/cancer-org/research/cancer-facts-and-statistics/breast-cancer-facts-and-figures/breast-cancer-facts-and-figures-2017-2018.pdf>
BCG. (2022). DICE® BCG’s DICE tool helps companies better predict project outcomes. Boston Consulting Group. https://www.bcg.com/capabilities/business-transformation/change-management/dice
Bovey, Wayne H., Andrew Hede.(2001) "Resistance to organisational change: the role of defence mechanisms." Journal of managerial psychology 16.7 pp: 534-548.
Bresnick J. (2016). Big Data, Artificial Intelligence, IoT May Change Healthcare in 2017. <https://healthitanalytics.com/news/big-data-artificial-intelligence-iot-may-change-healthcare-in-2017>
Bresnick J. (2018A). Machine learning, imaging analytics predict kidney function. <https://healthitanalytics.com/news/machine-learning-imaging-analytics-predict-kidney-function>.
Bresnick, J. (2018B). Top 12 ways artificial intelligence will impact healthcare. <https://healthitanalytics.com/news/top-12-ways-artificial-intelligence-will-impact-healthcare>
Carolina Lugo-Fagundo, Bert Vogelstein, Alan Yuille, Elliot K. Fishman. (2018). Deep Learning in Radiology: Now the Real Work Begins. Available at: <https://www.jacr.org/article/S1546-1440(17)30969-9/fulltext>
Dyche J. (2018). The promise of artificial intelligence in diagnosing illness. <https://www.cio.com/article/3305951/the-promise-of-artificial-intelligence-in-diagnosing-illness.html>
Frost & Sullivan (2016) Frost & Sullivan From $600 M to $6 billion, artificial intelligence systems poised for dramatic market expansion in healthcare. <https://ww2.frost.com/news/press-releases/600-m-6-billion-artificial-intelligence-systems-poised-dramatic-market-expansion-healthcare>
Griffiths S. (2016). This AI software can tell if you're at risk from cancer before symptoms appear. <https://www.wired.co.uk/article/cancer-risk-ai-mammograms>
Hsieh, P. (2017a). AI in medicine: rise of the machines. https://www.forbes.com/sites/paulhsieh/2017/04/30/ai-in-medicine-rise-of-the-machines/>
Hultman, K. (1979), The Path of Least Resistance, Learning Concepts, Denton, TX.
Kakar K. (2017). AI trends that are disrupting healthcare industry. <https://insights.daffodilsw.com/blog/4-ai-trends-that-are-disrupting-healthcare-industry>
Kent J. (2018). How artificial intelligence is changing radiology, pathology. <https://healthitanalytics.com/news/how-artificial-intelligence-is-changing-radiology-pathology>
Kooi, T., Litjens, G., van Ginneken, B., Gubern-Mérida, A., Sánchez, C. I., Mann, R., den Heeten, A., & Karssemeijer, N. (2017). Large scale deep learning for computer aided detection of mammographic lesions. Medical image analysis, 35, 303–312. https://doi.org/10.1016/j.media.2016.07.007
Kotter, J.P. and Schlesinger, L.A. (1979), ``Choosing strategies for change’’, Harvard Business Review, March-April, pp. 106-14.
Krittanawong C. (2018). The rise of artificial intelligence and the uncertain future for physicians. Available at: <https://www.ejinme.com/article/S0953-6205(17)30261-3/fulltext>
Liew, C. (2018). The future of radiology augmented with Artiﬁcial Intelligence: A strategy for success. Available at: <https://www.researchgate.net/publication/323844644_The_future_of_Radiology_augmented_with_Artificial_Intelligence_a_strategy_for_success>
Liu, R., Rong, B., Tu, P., Tang, Y., Song, W., Toyos, R., Toyos, M., & Yan, X. (2017). Analysis of Cytokine Levels in Tears and Clinical Correlations After Intense Pulsed Light Treating Meibomian Gland Dysfunction. American journal of ophthalmology, 183, 81–90. https://doi.org/10.1016/j.ajo.2017.08.021
Michael Recht, R. Nick Brya. (2017). Artificial intelligence: threat or boon to radiologists? Journal of American College of Radiology. 2017;14(11):1476–1480.
Milton, C., Entrekin, L. and Stening, B. (1984), Organisational Behaviour in Australia, Prentice Hall, Sydney,p.480
Olson, E.E. (1990), ``The transcendent function in organizational change’’, The Journal of Applied Behavioural Science, Vol. 26 No. 1, pp. 69-81.
Pearl R. (2018). Artificial intelligence in healthcare: separating reality from hype. Available at: <https://www.forbes.com/sites/robertpearl/2018/03/13/artificial-intelligence-in-healthcare/#71db2fc61d75>
Pearson, D. (2017). Artificial intelligence in radiology: the game-changer on everyone’s mind
Pratt, MK. (2018). Artificial intelligence in primary care. <https://www.medicaleconomics.com/business/artificial-intelligence-primary-care Medical economics. 2018>
Ruiz Hidalgo, I., Rozema, J. J., Saad, A., Gatinel, D., Rodriguez, P., Zakaria, N., & Koppen, C. (2017). Validation of an Objective Keratoconus Detection System Implemented in a Scheimpflug Tomographer and Comparison with Other Methods. Cornea, 36(6), 689–695. https://doi.org/10.1097/ICO.0000000000001194
Sappin E. (2018). 4 ways AI could help shape the future of medicine. <https://venturebeat.com/2018/02/20/4-ways-ai-could-help-shape-the-future-of-medicine/>
Schlegl T, Waldstein SM, Bogunovic H, Endstraßer F, Sadeghipour A, Philip AM, Podkowinski D, Gerendas BS, Langs G, Schmidt-Erfurth U. (2018). Fully automated detection and quantification of macular fluid in OCT using deep learning. Ophthalmology. 2018;125:549–558.
Singh, S. (2018). Artificial Intelligence and the internet of things in healthcare. <https://thejournalofmhealth.com/artificial-intelligence-and-the-internet-of-things-in-healthcare/> ;The Journal of Mhealth. 2018
Tahir, U. (2020). What is DICE Framework in Change Management?. Change Management Insights. https://changemanagementinsight.com/dice-framework/
Timothy J. W. Dawes, Antonio de Marvao, Wenzhe Shi, Tristan Fletcher, Geoffrey M. J. Watson, John Wharton, Christopher J. Rhodes, Luke S. G. E. Howard, J. Simon R. Gibbs, Daniel Rueckert, Stuart A. Cook, Martin R. Wilkins, Declan P. O’Regan. (2017). Machine Learning of Three-dimensional Right Ventricular Motion Enables Outcome Prediction in Pulmonary Hypertension: A Cardiac MR Imaging Study. Available at: <https://pubs.rsna.org/doi/full/10.1148/radiol.2016161315>
Waddell, D., & Sohal, A. S. (1998). Resistance: a constructive tool for change management. Management decision, 36(8), 543-548.
Weng SF, Reps J, Kai J, Garibaldi JM, Qureshi N. (2017). Can machine-learning improve cardiovascular risk prediction using routine clinical data? PLOS ONE. 2017;12(4):2017.
Wottawa CR, Genovese B, Nowroozi BN, Hart SD, Bisley JW, Grundfest WS, Dutson EP. (2016). Evaluating tactile feedback in robotic surgery for potential clinical application using an animal model. Surgical Endoscopy. 2016;30(8):3198–3209.

Table I. Interviewees responses.

Responses Distribution
Scale	None	Marginal	Moderate	Notable	Substantial
Interviewees Responses	-	2	1	2	5

Table II. Hypothesis one-sample statistics.

One-Sample Statistics
	N	Mean	Std. Deviation	Std. Error Mean
How much impact would AI have on doctor’s redundancy?	10	4.0000	1.24722	.39441

Table III. Hypothesis one-sample test.

One-Sample Test
	Test Value = 1
	t	df		Mean Difference	95% Confidence Interval of the Difference
			Sig. (2-tailed)		Lower	Upper
How much impact would AI have on doctor’s redundancy?	7.606	9	.001	3.00000	2.1078	3.8922

Download PDF

Version 1

posted

You are reading this latest preprint version

Bridging the Gap: Assessing the Integration of Artificial Intelligence in Healthcare for Improved Efficiency and Doctor Adaptability.

Status:

Version 1

Abstract

Overview of Artificial Intelligence in Health

Methodology

Discussion

Artificial Intelligence in Modern Medicine:

Artificial Intelligence in Diagnostic Radiology:

Artificial Intelligence in Pathology:

Artificial Intelligence in Ophthalmology:

Artificial Intelligence in Cardiology:

Future of Artificial Intelligence in Healthcare:

Transforming Medical Care Through Artificial Intelligence:

The Impact of Artificial Intelligence on Physicians:

Dice Scoring System:

Conclusions

Declarations

References

Tables

Status:

Version 1