COVID-19: An Algorithmic Model for Critical Medical Resource Rationing in a Public Health Emergency
Lead Research Organisation:
Durham University
Department Name: Management and Marketing
Abstract
The aim of the project is to develop an algorithmic model that calibrates a dynamic index for patient priority by addressing the shortcomings of the current allocation protocols of scarce medical resources.
The total number of confirmed Covid-19 deaths in the UK has already passed the 45,000 mark. Such a horrific number of deaths is partly attributable to the shortage of PPE, medical staff, and ICU beds in the early stages of UK pandemic. For a second wave of Covid-19 likely in the winter when the healthcare system is most stretched, scientists have estimated that the UK could see about 120,000 new coronavirus deaths.
To achieve the greatest good for the greatest number of patients, it is essential to have in place ethically and clinically sound policies on the allocation of scarce resources. Existing triage guidelines determine patient priority based on several attributes, including the illness severity and the near-term prognosis after discharge. They focus on individual patients but ignore the overall mixture of current patient profiles and the uncertainty in the number of patients who become critical ill over time. Previous research has shown that such frameworks could lead to preventative deaths and inefficient usage of scarce resources. We aim to address these limitations in this project via the development of an algorithmic model that calibrates a dynamic index (priority). Its performance is to be compared against the benchmarks via an empirical study using anonymised data of Covid-19 patients collected by Public Health England.
The total number of confirmed Covid-19 deaths in the UK has already passed the 45,000 mark. Such a horrific number of deaths is partly attributable to the shortage of PPE, medical staff, and ICU beds in the early stages of UK pandemic. For a second wave of Covid-19 likely in the winter when the healthcare system is most stretched, scientists have estimated that the UK could see about 120,000 new coronavirus deaths.
To achieve the greatest good for the greatest number of patients, it is essential to have in place ethically and clinically sound policies on the allocation of scarce resources. Existing triage guidelines determine patient priority based on several attributes, including the illness severity and the near-term prognosis after discharge. They focus on individual patients but ignore the overall mixture of current patient profiles and the uncertainty in the number of patients who become critical ill over time. Previous research has shown that such frameworks could lead to preventative deaths and inefficient usage of scarce resources. We aim to address these limitations in this project via the development of an algorithmic model that calibrates a dynamic index (priority). Its performance is to be compared against the benchmarks via an empirical study using anonymised data of Covid-19 patients collected by Public Health England.
| Description | At the beginning of the pandemic the disease was spreading rapidly, the number of deaths was climbing, and hospital, especially ICU was overwhelmingly approaching to its capacity. Such a strain on the existing health system creates the demand for rationing critical resource including ICU beds, medical equipment and interventions by employing modelling and simulation approaches. The existing prioritization/rationing policies developed are underpinned by the four fundamental values as reported in the leading journals in the field: "maximizing the benefits produced by scarce resources, treating people equally, promoting and rewarding instrumental value, and giving priority to the worst off. 24-29" In practice, when triage criteria are applied to individual patients at the time of ICU admission, their priorities are determined mainly by individual attributes based on single or some of those values, including the illness severity and the near-term prognosis after discharge. Nevertheless, they do not consider the dynamic illness trajectory of individual patients over the duration of treatment in ICU. Further, they ignore the overall mixture of current patient profiles and the uncertainty in the number of patients who become critical ill over time. Above all, the existing triage criteria do not adopt and operationalise all these values simultaneously, leading to the challenges that undermine the principle of fairness in resource rationing. We aim to address these limitations in this project via the development of an algorithmic model that prescribes patient priority. By using the multivalue ethical framework in developing rationing policies, our modelling approach stems from the principle of fairness in resource rationing. We are the first study that operationalises multivalue ethical framework that is widely accepted by the triage literature into an algorithmic model. Our results show that all four fundamental values are manifested in our study by applying the rationing policies developed. Our rationing policies performing better than other alternative allocation/prioritization heuristics at the most of times. The performance in benefit maximization is robust under a variation of experiments via both numerical and simulation studies. Similar percentage of patients in each category are admitted or rejected by applying our triage policies developed. Little imparity is observed across categories with similar prognosis in obtaining access to intensive care. By varying the discretion applied to early discharges, we show that as the penalty increases, our policy will eventually converge to FCFS that literally imply no triage is applied to patients. However, the value range of the penalty for early discharge where our policy outperforms others is wide enough. The ratios of the sickest/others and younger/others of those being admitted are both greater than 1 although close to 1. Nevertheless, among all the patients, the sickest from the younger group has been given the highest priority over others by applying our rationing policy at admission. |
| Exploitation Route | This project was carried out with the intention of preparing policy makers for the worst in a crisis. When the demand overwhelms the capacity and starts to compete for resources, it is essential to have rationing policies that are fair and speaks multifaceted ethical values in place. To the best of our knowledge, we are the first study that operationalises multivalue ethical framework that is widely accepted by the triage literature into an algorithmic model. Our results show that all four fundamental values are manifested in our study by applying the rationing policies developed. |
| Sectors | Healthcare |
| Description | This project was carried out with the intention of preparing policy makers for the worst in a crisis. When the demand overwhelms the capacity and starts to compete for resources, it is essential to have rationing policies that are fair and speaks multifaceted ethical values in place. To the best of our knowledge, we are the first study that operationalises multivalue ethical framework that is widely accepted by the triage literature into an algorithmic model. Our results show that all four fundamental values are manifested in our study by applying the rationing policies developed. From the open dialogues with health professionals, we learnt that UK ICU units have never reached a breaking point where two or more critical patients have had to compete for one bed. Nevertheless, our research is best placed to shed light on how to come up with effective rationing policies when such unfortunate scenarios do happen. Our results have shown that "interrupt strategy" if applied to ICU triage (which allows earlier discharge of an existing patient) would have provided larger incremental gains in terms of aggregated life-years saved during the pandemic. The findings have been disseminated to healthcare executives via presentation and workshops. To disseminate results and to exercise an influence on triage policy further, we presented our findings to our NHS partner, a group of intensive care consultants and clinicians in particular who will have to make triage decisions under those dreadful situations, inviting discussions on how likely our policies developed based on algorithmic models could be implemented in practice. We will take things further from there by attending and presenting our findings at conferences of health professional and policy makers. The nature of our modelling approach described above echoes the four fundamental values underlying the principle of fairness in rationing critical care resources, which are operationalized in the following ways: • Maximizing the benefits produced by scarce resources is explicitly embedded in our objective function aiming at minimising the total life-year lost over the given time horizon at an ICU. Further, at an individual patient level, higher relative benefit from their ICU stays as well as shorter length of stay play an important role in deriving prioritization policies. • Treating people equally is operationalized in our models by allowing early discharge in the sense that every critically patient is considered in the same way regardless of age, gender, prognosis, arriving earlier or later. For each critically ill patient at admission, not only must a decision made whether to admit or not, but also the severity level at which the patient will be discharged. • Promoting and rewarding instrumental value is understood by exercising discretion to particular groups of patients due to their profession or special contribution they made. In our study, an early discharge cost is applied to differentiate the existing patients as against the new arrivals. The same calibration can be readily tailored to address the discretion required for a particular group of patients in prioritization decisions without violating the overall aim of benefit maximization. • Giving priority to the worst off could be understood as giving priority either to the sickest or to younger people who will have lived the shortest lives if they die untreated. Our model has incorporated the very principle by using life year remaining instead of the number of lives saved in our overarching objective to give favourable consideration towards younger people. As the sickest are usually associated the lower relative benefit in proportion to the length of stay, they might be among those who are first early discharged. However, by increasing the early discharge cost, we provide a lever to ICU clinicians as to what extent they would give priority to the existing sickest patients. Up to our knowledge, we are the first to propose a cost parameter h associated with penalizing early discharges in the triage literature. By adjusting h, our proposed policy can be calibrated to strike a balance between minimizing life-years lost and managing the rate of early discharges. In consequence, it helps reconcile the inherent divergence between Maximizing the benefits produced by scarce resources and Giving priority to the worst off, between Treating people equally and promoting and rewarding instrumental value (for existing patients in our case). With an increased h, our algorithmic fairness tilts towards giving priorities to the existing sickest and applying more discretion to existing ICU patients, whereas maximizing overall benefits by minimising life-years lost, giving priority to younger patients and treat people equally are more achievable with a reduced h. |
| Sector | Healthcare |
| Impact Types | Policy & public services |
| Title | An Algorithmic Model for Critical Medical Resource Rationing in a Public Health Emergency |
| Description | • The modelling of the problem as discrete MDP; • The modelling of patient progression during their stay in ICU as Markov Chains with two absorbing states; • Two decomposition methods proposed to derive index policies for patient prioritization; • Inputs and engagement from ICU clinicians and health professionals in estimating probabilities of patients mortalities if they were denied admission to ICU; • Both numerical and simulation study show that our index policy developed works quite well as compared to the optimal solutions and other alternative triage policies. • Both programming codes in R and research findings are still in progress. When completed, it will be shared via open source platform such as Github. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2021 |
| Provided To Others? | No |
| Impact | Many existing Covid-19 projects focus on the prediction of the demand for scarce resources such as ICU beds and ventilators. For example, the system Adjutorium developed by Professor Mihaela van der Schaar at Cambridge uses artificial intelligence to predict how Covid-19 impacts resource needs. However, they do not consider an important question after the prediction: how the scarce resources should be allocated if they are overwhelmed. This is indeed the aim of this project. The project aims to develop an algorithmic model and evaluate its potential benefits in support of rationing decisions as against multivalue ethical framework widely accepted in the literature for fairness in resource allocation. We are the first study that operationalises multivalue ethical framework that is widely accepted by the triage literature into an algorithmic model. Our results show that all four fundamental values are manifested in our study by applying the rationing policies developed. Our rationing policies performing better than other alternative allocation/prioritization heuristics at the most of times. The performance in benefit maximization is robust under a variation of experiments via both numerical and simulation studies. Similar percentage of patients in each category are admitted or rejected by applying our triage policies developed. Little imparity is observed across categories with similar prognosis in obtaining access to intensive care. By varying the discretion applied to early discharges, we show that as the penalty increases, our policy will eventually converge to FCFS that literally imply no triage is applied to patients. However, the value range of the penalty for early discharge where our policy outperforms others is wide enough. The ratios of the sickest/others and younger/others of those being admitted are both greater than 1 although close to 1. Nevertheless, among all the patients, the sickest from the younger group has been given the highest priority over others by applying our rationing policy at admission. |
| Description | Collaborative partnership with UHBW |
| Organisation | University Hospitals Bristol and Weston NHS Foundation Trust |
| Country | United Kingdom |
| Sector | Hospitals |
| PI Contribution | Through close engagement with ICU consultants at UHBW, we have collected the probabilities of survival if patients were rejected from ICU admissions via those experts who are working at the front line. Such probabilities are rarely recorded/estimated in the medical literature. This will enable clinician to quantify relative benefits gained from ICU admission of critical patients. |
| Collaborator Contribution | UHBW agreed to share anonymous ICU data of Covid patients admitted during the pandemic period with us for the project. |
| Impact | The collaboration is ongoing. As there is delay in data share agreement and data access, we are still working on the data analysis to establish a transition matrix for patient acuity progression during ICU stay. |
| Start Year | 2021 |
| Description | Maximising patient benefit from triage policies for scarce critical care resource allocation during periods of intense pressure: an empirically informed systems modelling and simulation study28 June 2023 - 28 June 2023 |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Professional Practitioners |
| Results and Impact | One of the signature online talk organized by Wolfson Research Institute of Health and Wellbeing, Durham |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://www.durham.ac.uk/research/institutes-and-centres/wolfson/about-us/events/maximising-patient-... |
| Description | Workshop for Healthcare Executives at City Bayes Business School |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | National |
| Primary Audience | Professional Practitioners |
| Results and Impact | 50 Healthcare executives attended for the workshop that consists of a presentation of the project and eliciting patient mortalities if their access to care were declined, which sparked questions and discussion afterwards. Interrupt strategy applied to ICU triage in the project was a new idea to many of audience, who have shown great interest and insightful comments from their relevant experience. |
| Year(s) Of Engagement Activity | 2021 |