Clinical-scientist-led transoesophageal echocardiography (TOE): using extended roles to improve the service

At the North West Anglia NHS Foundation Trust, we perform transoesophageal echocardiography (TOE), a semi-invasive diagnostic test using ultrasound for high-quality heart imaging. TOE allows accurate diagnosis of serious heart problems to support high-quality clinical decision-making about treatment pathways. The procedure can be lengthy and is traditionally performed by a consultant cardiologist, who typically has multiple commitments. This constrains patient access to TOE, leading to waits from referral to test, delaying treatment decisions. In this quality improvement project, we improved access by redesigning workforce roles. The clinical scientist, who had been supporting the consultant during TOE clinics, took on performing the procedure as the main operator. We used the Model for Improvement to develop this clinical-scientist-led service-delivery model, and then test and refine it. This increased capacity and frequency of TOE clinics, reducing waits and releasing around 2 days per month of consultant time. Over five plan-do-study-act cycles, we tested six changes/refinements. Our targets were to reduce the maximum waiting time for TOE to 3 working days for inpatients and to 14 working days for outpatients. We succeeded, achieving reductions in mean waiting times from 7.7 days to 3.0 days for inpatients and from 33.2 days to 8.3 days for outpatients. TOE requires intubation; when this fails, TOE is abandoned. We believe light (rather than heavy) sedation is helpful for this intubation. We reduced sedation levels (from a median of 3 mg of midazolam to 1.5 mg) and, as a secondary outcome of this project, reduced the intubation failure rate from 13% to 0% (over 32 postchange patients). Following this project, our TOE service is usually performed by a clinical scientist in echocardiography who has British Society of Echocardiography TOE accreditation and advanced training. We have sustained the improved performance and demonstrated the value of enhanced roles for clinical scientists.


Appendix B -Statistical Analysis
The QI project involved changing from a 'Pre' regime, under which the Consultant Cardiologist decided the volume of sedative and operated the TOE probe, to the 'Post' regime, under which the Clinical Scientist took over both tasks, so there are two independent variables Sedation and Operator.The outcome is an intubation Success or Failure.In this appendix we apply some further statistical tools, but with great caution since i) we have a limited amount of data and ii) it is observation data (from a natural rather than designed experiment).The data are too sparse to look at both Operator and Sedation.We can though consider the association of the Sedation with the outcome, assuming no effect of Operator and no systematic difference between the patients under the two regimes.

Impact of Sedation
Note: Sedation was administered in units of 0.5mg.The X-Y graph below uses jittering (adding a small amount of random noise to the data display) to avoid overprinting; the histogram uses 'dodging' (showing the split data offset) also to avoid overprinting.

Analysis of Volume of Sedation
We can see that

Analysis of Outcome by Volume of Sedation
And the probit model: where Φ is the cumulative distribution function of the standard normal distribution.
Give almost identical results, as shown in Figure B1 (upper), with • Sedation being statistically significant (p=.015) in both.
BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) BMJ Open Qual doi: 10.1136/bmjoq-2023-002268 :e002268.12 2023; BMJ Open Qual , et al.

Kaye N
Using cross-validation [1] (with k=2 folds, because the dataset is fairly small, and 20 repeats), using the caret package, [2] gives Cohen's κ (kappa) of around .72 with standard deviations of around .17 (Monte Carlo techniques are used, so results differ slightly each time the procedure is run).
κ is a measure of correlation between raters (e.g.models vs observed) robust to imbalanced groups.Cohen suggested values of .61 to .80 indicate substantial agreement; a more recent suggestion is that .60 to .79 be regarded as moderate agreement.[3] • We can therefor regard the logit or probit models as moderately robust (even with this small amount of data) We might typically locate Sedation ≤ 4.5mg (i.e. as a threshold (P(Success) remains above 50%) but, given the impact on the patient and waste of resources from a failure (abandonment of the procedure), we might want to suggest a threshold with a much higher probability of success.
Experimenting with a conditional inference classification and regression tree, [4] suggests a threshold of Sedation ≤ 3.5mg, splitting our 58 datapoints with that level of Sedation and 100% success, versus the 11 above where success was 7/11 = 64%.
These analyses suggest this small study • does support the hypothesis that Sedation volume is indeed associated with intubation success, • and Sedation > 3.5mg should be regarded with caution.

Figure 2
Figure 2 in the main paper shows both the volume of sedative and the outcome (intubation success of failure) across the project.

•
the Sedation does tend to be different by regime (Pre vs Post) BMJ Publishing Group Limited (BMJ) disclaims all liability and responsibility arising from any reliance Supplemental material placed on this supplemental material which has been supplied by the author(s) BMJ Open Qual doi: 10.1136/bmjoq-2023-002268 :e002268.12 2023; BMJ Open Qual , et al.Kaye N in Table B1 and the histogram in B2 (and the Mann-Whitney test give p < .005)-under the Post regime less Sedation tended to be administered (by the Clinical Scientist) than in the Pre regime (by the Consultant).The medians are 1.5mg vs 3.0mg.

Figure
Figure B1 upper: scatterplot with probit and logit fitted curves; lower: histogram (note: data are jittered or dodged around the actual values which are multiples of 0.5mg) Attempting binary regression to model the outcome (Success / Fail) from the Sedation, by fitting the logit model (logistic regression):