POLSKI
Introduction
In 2018, Jonathan Bush reported in the Harvard Business Review that approximately 14% of healthcare spending, which amounted to $91 billion, was wasted in the US as a result of ineffective or excessive administrative activities often related to duplicate or multiple document circulation. The excess of bureaucratic procedures is touted as the „sewage” of modern medicine [1]. This indicates that traditional interconnections of bureaucratic activities in health care do not work. The use of artificial intelligence (AI) can enhance logistics and optimization of procedures leading to less effort and costs in the long run. Additionally, one of the basic results of these activities, achieved due to big data analysis, could be setting directions of cash flows, streamlining the management of existing funds, and reducing opportunity costs [2,3]. It should be appreciated that each doctor, who comes into contact with a patient, is exposed to a pile of information, often exceeding the perceptual capability. Excess information becomes a „noise bottleneck” hindering effective actions and generating unnecessary costs causing cognitive and neurotic disorders [4,5].
The use of AI software to develop medical files appears an essential step in reducing the paperwork burden for healthcare professionals who assist patients. Such trials, albeit being rapidly on the rise, are not yet the binding standards [6]. Therefore, this article aims to evaluate the effectiveness of several ideation models, with particular attentiveness to artificial intelligence (AI), for optimizing hospital workflow in a single clinical ward. Models included the development and processing of medical files, treatment, and personnel and service costs referring to patients with chronic musculoskeletal disorders or trauma. The primary outcome was the influence of AI on healthcare human resources and secondary outcomes dealt with medical facility’s running expanses.
Material and methods
This study was performed in accordance with the 1964 Helsinki Declaration and its later amendments, national legislation, and institutional requirements. It did not involve any direct contact with patients. Thus, ethics and consent-seeking requirements were waived by the institutional research review committee.
Nine sub-models were considered including an average volume of medical files and reduced or increased volumes, with full or partial use of AI, and the lack thereof. The models were generated using Microsoft Office Excel. The following assumptions were made: 1/ a 20-bed ward employing six doctors, 12 nurses, and unlimited resources of non-medical secretaries to process medical files; 2/ doctors and nurses worked 168 hours/month, with 80% of worktime excluding official breaks and holidays; 3/ hourly staff remuneration conformed to the Agency for Health Technology Assessment and the Tariff System in Poland used in orthopedic wards. Revenue and earning margins (revenue minus costs) were estimated at $1000 per patient. Different scenarios of assumptions are presented in the tables below.
Initially, we assumed that health care in the case of each patient involves a certain amount of doctor, nurse, and secretary’s time. This time was divided between treatment, fixed in the model, and that required for developing medical files, which was variable depending on the file volume (Table 2). The influence on the time involvement of human resources of the following three volumes of medical files was assumed: average commonly created volume and increased or decreased volume. The probability of occurrence of each was assumed at 40%, 50%, and 10%, respectively (Table 3).
Table 1
Human resources - availability of time and costs.
| Resources | Monthly time availability (h) | Hourly wages ($US) |
|---|---|---|
| Doctors | 806.4 | 20 |
| Nurses | 1612.8 | 10 |
| Secretaries | flexible | 7 |
Table 2
Time involvement of human resources.
| Resources | Treatment (h/patient) | Medical files (h/patient) | Total time (h) |
|---|---|---|---|
| Doctors | 4.5 | 1.5 | 6.0 |
| Nurses | 10.0 | 2.0 | 12.0 |
| Secretaries | 0.0 | 2.0 | 2.0 |
Table 3
Time involvement of human resources by medical file volumes.
| Resources | Average file volume (h/patient) | Increased file volume (h/patient) | Decreased file volume (h/patient) |
|---|---|---|---|
| Doctors | 1.5 | +0.5 | -0.5 |
| Nurses | 2.0 | +0.5 | -0.5 |
| Secretaries | 2.0 | Unchanged | Unchanged |
Finally, we evaluated the following three scenarios of implementation of AI for developing medical files: (a) lack of AI use, (b) partial implementation of selected AI features, and (c) full-scale AI implementation (Table 4). The monthly costs of these activities were estimated at $0, $500, and $1000, respectively.
Table 4
Time involvement of human resources by the use of artificial intelligence (AI).
| Resources | No AI (h/patient) | Partial AI use (h/patient) | Full-scale AI use (h/patient) |
|---|---|---|---|
| Doctors | 1.5 | -0.5 | -1.0 |
| Nurses | 2.0 | -0.5 | -1.0 |
| Secretaries | 2.0 | Unchanged | Unchanged |
Overall, we evaluated nine scenarios depending on the volume of medical files and the use of AI. They were enumerated as follows: 1a (average volume of files, no AI), 1b (average volume of files, selected use of AI features), 1c (average volume of files, full-scale use of AI); 2a (increased volume of files, no AI), 2b (increased volume of files, selected use of AI features), 2c (increased volume of files, full-scale use of AI); and 3a (decreased volume of files, no AI), 3b (decreased volume of files, selected AI features), 3c (decreased volume of files, full-scale use of AI). All calculations were performed using Microsoft Office Excel.
Indirect costs were estimated based on a human capital approach using GDP per working person. Costs were rounded off to the full US dollar. The assumptions made were as follows:
20% of people treated were professionally active and lost 10% of work time waiting for healthcare services;
In 2022, the monthly GDP per capita was $17740 in Poland. We assumed 22 workdays in a month, amounting to daily GDP per capita of $805.
At the onset, a line of 134 patients was assumed, waiting for healthcare services, i.e., the maximum hospital ward’s capacity (scenario 3c).
Costs were calculated for full months of waiting time.
Results
Financial consequences of each modeling scenario included the following: direct costs of personnel involved in developing medical files, costs of AI whenever appropriate, and changes in the contribution margin compared with scenario 1a, taken as a starting point, which could be negative or positive. Financial consequences ranged from $13144 to $43364 per scenario, which comes to about 3.3-fold differences (Tables 5-13).
Table 5
Costs for modeling scenario 1a (average volume of medical files, no AI).
Table 6
Costs for modeling scenario 1b (average volume of medical files, use of selected AI features).
Table 7
Costs for modeling scenario in 1c (average volume of medical files, use of full-scale AI features).
Table 8
Costs in modeling scenario 2a (increased volume of medical files, no AI)
Table 9
Costs in modeling scenario 2b (increased volume of medical files, use of selected AI features).
Table 10
Costs in modeling scenario 2c (increased volume of files, use of full-scale AI features)
Table 11
Costs in modeling scenario 3a (decreased volume of medical files, no AI).
Table 12
Costs in modeling scenario 3b (decreased volume of medical files, use of selected AI features).
Table 13
Costs in modeling scenario 3c (decreased volume of medical files, use of full-scale AI features).
The optimal solution for the hospital was a model in which the volume of developed medical files was decreased and AI was used full-scale (scenario 3c). Although individual scenarios did not differ significantly in direct costs, relieving the limited resources, i.e., doctors and nurses, from administrative processes allowed for increases in the number of patients admitted and contribution margins. In some modeling scenarios, free capacity remained after maximizing the number of patients, which could generate further value by engagement in other processes, e.g., consultations.
Depending on the scenario, indirect costs also changed depending on the number of people awaiting hospitalization (Table 5). This calculation was made just for one month, but the difference between demand and availability would increase in the long run. The lowest social costs were incurred for scenarios 3b and 3c where AI was fully used with a decreased volume of medical, followed by 1c where AI was used with an unchanged volume of files. Social costs were invariably lower for scenarios with the use of AI compared with other variants (Table 14).
Table 14
Monthly indirect costs.
The estimated hospital fiscal benefits were calculated assuming access to full and proper information. The choice of a volume of medical files was based on regulations beyond the hospital’s control. The probability of occurrence of each modeling scenario amounted to 40%, 50%, and 10% for the average, increased, and decreased volume of medical files, respectively. The fiscal losses varied, being the greatest in Scenario 2a and smallest in Scenario 3c (Table 15).
Discussion
In this study, we came up with models for optimizing hospital workflow in a single orthopedic ward by employing a range of AI uses, full-scale, partial, and no AI features. The models included a varied volume of medical files to be developed, treatments, and personnel and service costs and referred to the long-term patient care typical for chronic neuromusculoskeletal disorders or traumas. The results were compared from the standpoint of expected fiscal provisions and professional healthcare benefits. The estimated valuation of expanses indicated that the bigger the number of patients, the higher the revenue, the smaller the total costs, and the higher the earning margins. Further, the modeling showed that the most efficient and least pricey, with more than three-fold savings, scenario referred to the presence of a reduced volume of medical files and a full-scale use of AI features. The hospital costs incurred in this modeling scenario and those in the worst case of an increased volume of files and no AI use amounted to $13,144 and $43,364, respectively, a significantly more than 3-fold difference.
Implementation of AI in a medical facility decreases the involvement of staff in other than strictly medical activities, which is an all-too-often complaint appearing while surveying healthcare workers. AI use generates financial benefits for the facility by relieving and redirecting medical professionals to clinical processes [7]. The benefits depend on the costs incurred by a given facility and savings of time otherwise used for administrative work before the commencement of AI [8,9]. There is an added-on psychological value consisting of increased confidence and coherence in clinical undertakings often requiring time and attention to be thoroughly thought out to achieve optimal performance [10,11]. That all is consistent with other industries where robotization and automation take over repetitive, simple processes, allowing qualified human resources to be directed to activities that generate the greatest value [12,13].
The main factors limiting the spread of AI tools in medical facilities are legal and financial regulations. To this end, the lack of unified liability regulations for the content generated by AI comes to the fore in Poland as well as the EU. Currently, full responsibility rests with medical staff who undersign the files developed [14], which hampers the full-fledged use of support offered by new technologies. For instance, AI provides transcriptions of file contents dictated and approved by doctors or nurses. Other AI capabilities having to do with the elaboration of test results, diagnostics, procedures, pharmacotherapies, and information collected by health information systems cannot be legitimately used. Thus, AI is a time saver for medical staff and improves workflow, but, currently, it does not relieve the staff from other time-established clinical responsibilities. Besides, AI algorithms undergo steady modifications and improvements, the clinical knowledge of which leaves a lot to be desired. The benefits of implementing AI also are desirable from a social perspective, given medical staff shortages. In Poland, there were 3.3 doctors and 5.1 nurses per 1000 inhabitants in 2017, with the EU averages of 4.0 and 8.3, respectively.
Additionally, the number of patients requiring long–term nursing care due to multiple chronic conditions (multimorbidity) increased sharply the world over in recent years, reaching an estimated 50 million people in Europe with prospects of further increases in the years to come [15,16]. And Poland is no exception in the trend [17]. This shows the scale of the problem and justifies actions to engage as many as feasible healthcare professionals in clinical management and away from administrative activities by broad use of the sprouting AI algorithms.
In the present study, cost-benefit modeling included both costs related to direct patient management like therapy, diagnostics, or chronic care and also those incurred for the implementation of new technologies. The costs of implementing AI features were presented on an accrual basis, i.e., given the depreciation of the technology purchased over time. When assessing an investment and calculating its net expected value, it is necessary to analyze the costs and benefits incurred over the entire investment life cycle. The analysis has limitations inherent in ideation modeling, largely stemming from biases linked to assumptions made. In an attempt to minimize such biases, we presented nine crossover modeling scenarios, with a triple range of both medical file volumes developed and AI features considered, always referenced to the same number of 134 patients at onset (scenario 1a) taken as control.
Notwithstanding the limitations above outlined, we conclude that the implementation of AI into medical file processing benefits health workers’ clinical manageability in the workplace and reduces the socioeconomic strain of health care. The results of this modeling exercise strongly advocate the implementation of AI in clinical practice pointing to fiscal advantages, likely turning into benefits for particularly chronic patient management and nursing.
