• No se han encontrado resultados

Escenario de simulación 1: Red GPON de 8 ONU

CAPÍTULO 2. ESCENARIOS DE SIMULACIÓN

2.3 Escenarios de Simulación

2.3.1 Escenario de simulación 1: Red GPON de 8 ONU

This chapter briefly summarizes the found answers to all research questions (Section 8.1), relates the found results and their practical implications (Section 8.2), which leads to recommended interventions (Section 8.3).

8.1 Research questions

What is the current situation in the ED in Almelo?

The average length of stay is 2 hours and 24 minutes. The triage norms are met approximately 55% of the time. Furthermore, the staff experiences periods of crowding, mostly during the day, but also quiet periods, mostly at night and early morning. Weekdays are experienced as more crowded than weekends.

What is known in literature on ED crowding?

Causes of ED crowding can be divided into causes related to input, throughput and output, as are most potential solutions. Potential solutions typically aim to make optimal use of the available capacity or to reduce LOS and/or waiting times. Most often, LOS or census are used as KPI.

Important stakeholders are the ED specialists and the ED manager, who value time to triage, waiting times and LOS as important indicators of quality of care. Of all prediction methods, linear regression suits this situation best.

What are the causes of the perceived periods of crowding at the ED?

Variables correlated to the crowding score are mainly census related variables. To the surprise of the personnel, besides LOS and arrival-to-triage, waiting times seem to be a less relevant cause of crowding. The time factor, number of departures and number of U1 patients have the strongest predictive value, marking them as important causes. The importance of the time-factor indicates that the differences in patient/staff ratio over the week influences the perception of crowding. How can perceived crowding be quantified?

Of the crowding score, LOS and census, only census is found to be objectively predictable using the available data. The average census does not correlate perfectly with the crowding scores given by the staff, but it is more objective and census data is relatively easily available. Therefore, census is used to quantify crowding when making predictions.

How can periods and amount of ED crowding be predicted?

Linear regression is used to predict crowding based on literature and the available data. All models satisfy the assumption tests of linear regression sufficiently to be used in practice, however, only the census model has sufficient reliability. When looking at census per hour, there is a clear pattern which shows an increase around 10:00, peaks around 14:00 and starts to decrease (though more slowly than the increase in the morning) from 18:00 onwards. There are also differences in number of patients between the days of the week. Including these patterns in the prediction model, using a time-factor, improves the prediction.

64 How can the amount and intensity of the periods of ED crowding be decreased, using the

predictions?

Five potential improvements are suggested:

- Integrated triage: if both a doctor and nurse are available, the first two steps in the patient care-path can be combined, reducing the total waiting time between steps

- Redirecting plaster patients to reduce the number of patients which visit the ED - Reducing pick-up time using better communication and agreements with wards - Using crowding predictions combined with thresholds to decide when to call extra

personnel, either internal or off-duty personnel. - Changing or adding personnel shifts

How can potential solutions be evaluated?

The interventions can be evaluated using an existing simulation model of the ED in Almelo. Before this can be done, however, the model is updated and revalidated. Besides updating assumptions and input parameters, the most important updates are changing the model to a continuous one

(simulating 24/7) and the implementation of the A-unit and B-unit. Besides this, changes are made to simulate the interventions. For instance, the prediction model is incorporated into the simulation model. Based on literature, stakeholder analysis, and the data analysis, it is decided to compare the interventions’ performance based on average LOS, while taking the LOS of the A-unit and B-unit into account.

What is the expected performance of the potential solutions?

The effects of the different experiments lead from a few percentages increase in LOS (shifting the nurse shifts backwards) to more than 20% reduction of LOS (combining integrated triage with reduced pick-up time among others).

8.2 Results in practice

The crowding score measurement resulted in some surprises: Wednesday is perceived as most crowded while it is not the day with the most patients, and the number of departures has a bigger effect on perceived crowding than the number of arrivals. Other important factors: the number of U1 patients and the time-factor, are more intuitive. The results indicate that the number of patients in the ED plays an important role but is not solely responsible for perceived crowding.

It is unfortunate that the crowding score prediction model is not reliable enough for practical use. Either more/different variables, or a smaller time-step (e.g., measurements per hour) are needed. In practice though, besides being far more reliable, the census model has some important advantages over a crowding score model. A census model is more objective, and the data is more easily available since it only requires census data as input. The census model’s accuracy is (very) good up to two hours in advance but, while it stays reasonable, reliability decreases after two hours. In practice this is not a problem as most interventions will be based on the predictions up to two hours in advance. As expected, integrated triage and reducing pick-up time reduce the LOS. Another expected, but nonetheless positive result, is the positive interaction effect between reducing pick-up time and using AMU (AOA) nurses instead of IC nurses when extra capacity is needed, as this combination is strongly considered. Redirecting the plaster patients on the other hand, does not result in the expected magnitude of LOS decrease. When compared to similar experiments in the sensitivity analysis, the reduction is very small. Redirecting plaster patients is likely to cost more effort than it reaps rewards, since redirecting patients to an inpatient plaster room will likely have a big effect on that plaster room, which is not used to non-elective patients.

65 The conclusion that adding nurse shifts is typically preferable over adding doctor shifts is favourable in practice, since the number of nurses is bigger than the number of doctors, making scheduling nurses slightly more flexible. The ED specialist especially, are hard to schedule due to scarcity. When implementing integrated triage, adding ED specialist capacity is preferable, indicating that the bottleneck of integrated triage is typically the doctor, not the nurse. This is not strange, but is difficult in practice due to the aforementioned scarcity.

The results indicate that adding extra personnel temporarily, using census predictions, has a bigger effect on LOS decrease than adding shifts of the same staff types, often using less capacity overall. This is a promising result for research in the area of combining crowding predictions with

interventions. In practice, it might be hard to implement though, as it requires a relatively high staff flexibility. Currently, it can be hard to find extra staff in case of crowding, and this extra staff typically stays a full shift if they come in when off-duty, which is typically not needed according to the

simulation model. To reach this level of flexibility, either back-up shifts, or a matching compensation system might be needed.

8.3 Recommendations

The following interventions are promising: - Integrated triage

- Reduced pick-up time, particularly when combined with calling AOA nurses based on the number of patients in the waiting room

- Using crowding thresholds to call in temporary extra personnel based on predictions instead of adding extra shifts

We advise to implement integrated triage, combined with a pick-up time reduction, using AOA- nurses which are called when the waiting room contains more than the threshold number of

patients (recommended 2) and calling an extra nurse when the predicted total number of patients in the ED (including waiting room) is bigger than 80% of the capacity, and both an extra nurse and an extra ED specialist when the predicted number of patients in the ED exceeds 90% of its capacity. This approach is preferred over adding extra standard shifts. However, it requires more flexibility of the personnel.

If this is infeasible, add early (10:00 – 16:00) nurse shifts throughout the week or an early nurse shift during the week and a late ED specialist shift (16:00 – 24:00) during the weekend.

When predicting crowding, use census as KPI, as it is more reliably predictable than both the crowding score and LOS, and is more objective than the crowding score.

Trying to decrease the number of plaster patients at the ED by redirecting them has a small effect on the ED LOS, while it might affect other parts of the hospital (which the patients are redirected to), thus is inadvisable. The effects of implementing integrated triage and reducing pick-up time are significantly bigger than the effects of adding capacity, both using crowding thresholds and by adding shifts. Therefore, it is advised to focus on these two interventions.

66

Documento similar