The clinical and cost effectiveness of a virtual fracture clinic service

Data collection: clinical (outcomes a-f)

Data regarding outpatient activity including attendances, discharges, non-attendances and time to first clinic appointment are generated directly from the study site’s electronic Patient Administrative System (PAS) and stored in the Trust’s Outpatient Commissioning Data Set, which is used primarily for monitoring and payment purposes. Relevant data was received from our Information Governance Team in the form of Excel pivot-tables (Microsoft Corporation, Albuquerque, New Mexico), and included monthly activity for all fracture clinics over the study period. No demographic data was available.

Data-collection: financial (outcome g)

National Tariff prices for the different types of fracture clinics were obtained from various NHS England websites,^19,20 whilst corresponding locally-agreed prices were verified by the Income and Contracts Manager within the Finance Team (Table I). In addition to agreeing higher tariff prices for all fracture clinic types (Table I), managers at the study site agreed with the CCG for the higher amount of £160 to be paid for all first follow-up appointments where patients were referred directly from VFC, in order to offset the anticipated drop in income resulting from fewer new face-to-face clinic referrals.

The Band 6 VFC Nurse salary was confirmed by our organisation’s Human Resources team to be £34 530 per annum in line with the national NHS pay scale, with additional London weighting giving a total of £39 013 per annum.²¹

Our VFC model

The ‘intervention’ was the introduction of VFCs on 1 December 2014, in response to a transformation Commissioning for Quality and Innovation (CQUIN) target from the local CCG.

CCGs are local NHS organisations, created by the Health and Social Care Act (2012). As more than 90% of patients’ NHS contact is with their GP, CCGs were created to enable GPs to buy the services that their patients need. All GP practices within a given geographical area work together as a CCG to buy services that they refer their patients on to, such as hospital, mental health and community services. CCGs are led by a governing body of elected local GPs supported by other clinicians and NHS managers.

The CQUIN scheme is a form of pay-for-performance, enabling CCGs to reward excellence by linking a proportion of healthcare providers’ income to the achievement of local quality improvement goals.

Ours is a lean VFC model, requiring the employment of one new full-time Band 6 VFC Nurse but with no additional staffing or overhead costs. VFCs run alongside traditional face-to-face clinics, with the consultant spending the first 90 minutes in VFC before joining his colleagues in face-to-face clinic. Management plans are communicated directly to the patient via telephone thereafter by the VFC Nurse.

Referral pathways and patient flow through fracture clinics at the study site are shown both pre-intervention (Fig. 1) and post-intervention (Fig. 2). Access to face-to-face clinic has been restricted to the on-call orthopaedic team, whilst ED, GP and MIU staff can book directly into VFC.

Fig. 1

Referral pathways to fracture clinics before the intervention. Emergency department (ED), GPs and minor injuries units (MIU) could directly book patients into face-to-face clinics in addition to the on-call orthopaedic team.

Fig. 2

Referral pathways to fracture clinics after the intervention. Emergency department (ED), GPs and minor injuries units (MIU) can book patients directly to VFC only, or refer to the on-call orthopaedic team which is the only route for booking into face-to-face clinic. Patients reviewed in VFC can subsequently be followed-up in face-to-face clinics.

Study design: interrupted time series (outcome measures a,c,e,f)

A useful tool for evaluating the effectiveness of healthcare interventions is ITS.^22,23 Data is collected at several time-points both before and after a discrete intervention, to determine whether observed changes in outcome are explained by secular trends in the data or are likely attributable to the intervention itself.²² It also allows for controlling for confounders such as seasonality,²³ and is preferable to case-control or cohort studies where the inclusion of a control group is unfeasible (such as in this instance where the whole population was subject to the new referral pathway).²⁴

Our data meets the requirements for ITS analysis as set out by Wagner et al²³ in 2002, with clearly defined pre and post-intervention periods, similar numbers of patients in both groups (n₁ = 8771, n₂ = 8345) and a continuous, repeated series of observations performed on the study population collected at 36 equal time-intervals (19 pre-intervention and 17 post-intervention).

Initially, summary and descriptive statistics were obtained to familiarise researchers with the data and to allow for a priori determination of a best-fit regression model - a level-and-slope-change model was chosen for all outcomes as follows: Y_t = β₀+β₁T+β₂X_t+β₃TX_t(where T = time since start of the study period (months), X_t = binary variable indicating pre-intervention (0) or post-intervention (1) period, Y_t = outcome at time t, β₀ = baseline level at T=0, β₁ = pre-intervention trend or slope, β₂ = level-change following intervention and β₃ = slope change following intervention).

Segmented regression analysis with least-squares (ordinary) fit was performed. The node between pre- and post-intervention segments was constrained to data-point 20 (December 2014; introduction of VFCs).

The presence of autocorrelation within the data was excluded by plotting the residuals and checking for normality (D’Agostino & Pearson test);^25,26 autocorrelations were thus not modelled. There was minimal seasonality upon preliminary analysis therefore no dummy variables were included in the regression model.

During the study period there were no additional changes to the format of fracture clinic services, therefore we have made the assumption that VFC was the sole intervention for this population. No patient-specific covariates could be identified as no demographic data were available.

Before-and-after comparisons

For outcome measures b and d we opted to perform simple before-and-after comparisons as data were non-continuous. Data was checked for normality using the D’Agostino & Pearson test^25,26 and parametric analyses performed appropriately.

Mean maximum clinician-time per patient in face-to-face clinic was calculated using monthly attendance figures (obtained from the Trust data-set) and information regarding the number, duration and staffing of these clinics, for both pre- and post-intervention periods. Number of clinics: six face-to-face clinics per week, every week, however they do not run on Bank Holidays of which there are eight per calendar year. Clinic duration: three hours (180 minutes). Staffing: three orthopaedic surgeons. Since the introduction of VFC the consultant spends 90 minutes in VFC before joining his colleagues in face-to-face clinic. The total number of face-to-face ‘clinician-minutes’ per clinic was therefore 540 minutes in the pre-intervention period and 450 minutes thereafter.

Calculations are shown in Table II Mean clinician-time per patient was calculated by dividing by the mean number of clinician-minutes per month by the actual number of patients attending face-to-face for that particular month; monthly figures were used to allow more statistically accurate comparisons to be made between pre and post-intervention periods.

Financial (outcome measure g)

Before-and-after comparison of the cost to the CCG for commissioning all outpatient fracture clinic services at the study site was made for the financial year prior to the intervention and the financial year after. This was achieved by simple numerical calculations based on the attendance figures for each clinic type (obtained in the Trust data-set) and the agreed tariff prices (Table I).

Statistical analyses

Data was stored in Microsoft Excel workbooks (version 15.20, Microsoft Corporation), and all numerical calculations performed using programmed formulae available in this software. Statistical analyses were performed in Graph Pad Prism version 7.01 (La Jolla, San Diego California).

Student’s t-test was the parametric test used for direct before-and-after comparisons as all data were found to be normally distributed, whilst segmented regression was performed to complete the ITS. All p-values reported are two-tailed, with an arbitrarily-assigned significance level of 0.05. All means are reported with standard deviation (sd).

Are VFCs clinically effective?: Number of new patients attending face-to-face clinic per month

ITS segmented regression for number of new patients seen in face-to-face clinic per month is shown in Figure 4. The post-intervention period is shaded grey, and the counterfactual scenario (expected scenario had the intervention not occurred, based on pre-existing data-trends) is indicated by a dashed line.

Fig. 4

Interrupted time series (ITS) segmented regression for number of new patients seen in face-to-face fracture clinic per month, showing a level-and-slope change at 20 months (December 2014, the intervention). Solid dark orange line, regression line for each segment; Dashed dark orange line, counterfactual scenario; light orange dotted line, difference between counterfactual and actual scenario at six months post-intervention; Grey shading, post-intervention period.

Pre-intervention, a mean 461.6, sd 61.63 patients were seen in face-to-face clinic per month with an upward trend indicating 4.9 additional patients seen month-on-month. Post-intervention, there is a clear change in both slope and level with a mean 140.4, sd 39.62 patients seen and a downward trend indicating 5.2 fewer patients seen each month compared with the previous one. At six months after intervention, there is a statistically significant decrease in the number of patients seen in face-to-face clinic of 27.8% compared to the counterfactual scenario (149.8 versus 539.4 patients per month, segmented regression p < 0.0001).

Clinical time per patient in face-to-face clinics

Pre-intervention, each patient saw a doctor for a mean maximum time of 13 minutes 44 seconds (sd 2 minutes 8 seconds) increasing post-intervention to 14 minutes 53 seconds (sd 1 minute 46 seconds), a statistically non-significant increase of one minute nine seconds (Student’s t-test, p = 0.0878).

Number of days to orthopaedic clinical review

ITS segmented regression for patient wait between referral and first orthopaedic review in clinic is shown in Figure 5. Pre-intervention, the mean wait was 10.9, sd 1.5 days compared with 5.2, sd 0.66 days post-intervention (Student’s t-test, p < 0.0001). The pre-intervention trend showed increasing wait-times, with a month-on-month increase of 0.1 days. This trend slowed post-intervention, now displaying a monthly increase of only 0.013 days.

Fig. 5

Interrupted time series (ITS) segmented regression for number of days to first clinical review by an orthopaedic doctor, showing a level-and-slope change at 20 months (December 2014, the intervention). Pre-intervention slope, 0.12; post-intervention slope, 0.013. Solid dark orange line, regression line for each segment; Dashed dark orange line, counterfactual scenario; light orange dotted line, difference between counterfactual and actual scenario at six months post-intervention; Grey shading, post-intervention period.

At six months post-intervention, the waiting time was 5.1 days but would have been expected to be 12.8 without the intervention; this is a significant decrease by 40.1% (segmented regression, p < 0.0001).

Percentage of patients assessed within the 72-hour target recommended in BOAST 7 guidelines

Pre-intervention, 5.1% (sd 0.49; 447 of 8771) of new patients referred to the fracture clinic had their first appointment within 72 hours, compared to a post-intervention figure of 46.4% (sd 3.29; 3873 of 8345); this is a statistically significant increase (Student’s t-test, p < 0.0001, Fig. 6).

Fig. 6

Graph of Student’s t-test results comparing the percentage of patients reviewed within the BOAST 7 72-hour target time-frame in the post-intervention vs pre-intervention groups. There was a significant increase from 5.1% (± 0.49) to 46.4% (± 3.29), Student’s t-test, p < 0.0001.

Number of patients discharged from first face-to-face appointment

Monthly discharge rates from VFC since its introduction are shown in Figure 7; mean discharge rate is 32.9%, ranging from 25.9% (January 2015) to 39.5% (April 2016).

Fig. 7

Graph of study site Virtual fracture clinic (VFC) discharge rates per month over the entire study period. Mean discharge rate is 32.9% (range 25.9 to 39.5%).

Figure 8 shows ITS segmented regression for monthly numbers of patients discharged from first face-to-face fracture clinic, which acts as a proxy for unnecessary referrals as patients did not require further orthopaedic input.

Fig. 8

Interrupted time series (ITS) segmented regression for number of patients per month discharged directly from their first face-to-face clinic. There is a clear level and slope change. Pre-intervention slope, +3.7; post-intervention slope, –1.5; Solid dark orange line, regression line for each segment; Dashed dark orange line, counterfactual scenario; light orange dotted line, difference between counterfactual and actual scenario at six months post-intervention; Grey shading, post-intervention period.

Mean number of direct discharges decreased significantly from 129.2 (sd 7.36) to 33.5 (sd 3.66) per month following the intervention (p < 0.0001, Student’s t-test). Pre-intervention there was an upward trend, with an increase of 3.7 discharges per month. Post-intervention, this trend has reversed with 1.5 fewer discharges month-on-month. At six months post-intervention, we would expect 188.2 direct discharges according to pre-intervention trends, therefore the observed figure of 38 discharges represents a 79.8% significant decrease (segmented regression, p < 0.0001).

Number of non-attendances at first face-to-face appointment

The mean absolute number of patients who did not attend their first face-to-face clinic appointment decreased from 60.5 (sd 2.68) to 14.8 (sd 1.48) per month after the intervention (p < 0.0001; Student’s t-test; Fig. 9). At six months post-intervention, there was a statistically significant 75.0% reduction in non-attendance figures between the observed values and those expected in the counterfactual scenario (14.9 versus 59.4; segmented regression, p < 0.0001).

Fig. 9

Interrupted time series (ITS) segmented regression of absolute number of non-attendances per month for face-to-face clinic. There is a clear level change. Pre-intervention slope, 0.063; post-intervention slope, 0.11; Solid dark orange line, regression line for each segment; Dashed dark orange line, counterfactual scenario; Grey shading, post-intervention period.

As each missed first face-to-face appointment costs the study site £160, these results equate to an £81 920 saving in the 12 months post-intervention compared with the previous 12 months, based on monthly non-attendance figures.

In terms of percentages, there was a reduction of 2.5% in non-attendance rates from first-attenders at face-to-face fracture clinic from 13.1% to 10.6% (Risk Ratio (RR) 0.81, 95% confidence interval (CI) 0.71 to 0.92; Odd Ratio (OR) 0.78; 95% CI 0.68 to 0.9) between the pre- and post-intervention periods.

Figure 10 illustrates the effect of the intervention on all parameters for which ITS analysis was used. A decrease in absolute figures for non-attendance, immediate discharge and number of face-to-face clinic appointments per month can be clearly seen, in addition to decreased wait times from referral to clinic review.

Fig. 10

Graphical representation of the changes in key clinical performance parameters over the study period (see legend). The solid dark orange line shows the intervention (virtual fracture clinic (VFC) introduction in December 2014).

Are VFCs cost-effective?: total spend by the local CCG

Calculations used to obtain the total cost to the local CCG for commissioning fracture clinic services at the study site in the financial year before and after the intervention are shown in Table III. These are based on the numbers of each clinic appointment-type and locally-agreed tariff prices (Table I). In the first year of VFC operation the study site also received the £62 500 incentive bonus from the CCG for having achieved their CQUIN target; this will not be paid in subsequent years.

Pre-intervention, the CCG paid the study site £1 466 666.72 for fracture clinic services whereas post-intervention it paid £1 336 781.05 plus the £62 500 bonus (total £1 399 281.05). Thus, in their first year of operation VFCs saved the CCG £67 385.67, whilst in future years this can be expected to rise to £129 885.67, assuming fracture clinic activity remains relatively constant.

The Trust’s overall balance was £106 398.67 less in the first year of VFCs than previously, and looks set in subsequent years to be £168 898.67 less compared to pre-intervention figures due to a combination of lower CCG spending as above, provision of the VFC nurse’s salary and no further CQUIN bonus payments.

Main findings

The main findings of this study are that VFCs have led to: Fewer referrals to face-to-face fracture clinic (Figs 3 and 4), improved waiting times from referral to first orthopaedic review in clinic, ensuring more timely management decisions (Figs 5 and 6), fewer unnecessary referrals (Fig. 8) and non-attenders (Fig. 9) meaning more patients were saved the time and expense of travelling to hospital unnecessarily and saving the Trust £81 920 over 12 months through reducing missed appointments, and substantial financial savings of an estimated £129 885.67 per annum for the local CCG (Table III)

The null hypothesis is therefore rejected for outcome measures a, c, d, e, f and g but accepted for b where no significant difference was detected. It can thus be postulated that this particular VFC model is both clinically and cost-effective.

Strengths and limitations

This study addresses the current ‘hot topic’ of VFCs, given the context of overburdened traditional fracture clinic services and recent calls by NICE and the British Orthopaedic Association for further research in this area.^1,18 It is a large-scale study, including 17 116 patients referred over a three year period, and is to our knowledge the only paper in England to investigate cost-effectiveness of a VFC model. It captures a whole-population group including all referral sources, thus selection bias is eliminated.

ITS is a robust quasi-experimental methodology for assessing the longitudinal effects of population-level healthcare interventions over a defined time-period, and is widely recognised as a preferred design where randomisation is unfeasible.^22-24 Cochrane have recommended ITS for examining the effect of such healthcare interventions.²⁷ Its strength lies in its ability to adjust for secular trends in pre-intervention data, biases within the data itself and in allowing easy visualisation of the intervention-effects.²³ As the effect-lag after the intervention is extremely short in this case, it is possible to see stark differences in the data around the intervention point.

One shortfall of our model is that no demographic data was available to test for covariance; if it was it may have been possible to include a control group.

In relation to the literature

As of August 2016, the databases PubMed, OVID, PubMed, and Google Scholar contained only seven papers including original research, audit or commentary on the topic of virtual or telephone clinics within the NHS,^12-17,28 six of these were from the Glasgow Royal Infirmary.^12-17

Compared with these publications, our study is larger and examines outcomes over a longer time-period: the largest paper from Glasgow reported on 6385 patients over a one year period.¹³ Ours is also the only study to use a more robust methodological approach than simple before-and-after analysis. However, we have not evaluated the impact of our VFC on local ED, MIU or GPs as the Glasgow group have done;^12,13 this may be an area for further research.

We found higher discharge rates from our VFC than those of previous papers, which may be a symptom of our specific VFC model.^12,16 The study is generally limited in that it only applies to our specific VFC model within a particular population demographic. This is true for all published work on VFCs, however, and more publications are required in order for meta-analyses to be carried out.

Patient experience and satisfaction were not investigated as part of this study, nor were patient complaints or complications associated with the service redesign. However, upon inspecting our hospital-level data for fracture-clinic-related complaints there has been a slight reduction in these after VFC was introduced. There was only one early complaint related to VFC from all 5958 patients in the study, where the patient had concerns that no orthopaedic specialists had been involved in formulating his management plan. In line with the Glasgow group’s findings there have been no identified serious untoward incidents, episodes of patient harm or medico-legal actions since the intervention.¹⁷

Implications of this study

There are potential wider societal benefits, economical and environmental advantages of reducing numbers of face-to-face clinic attendances. Each patient has their own carbon footprint, and many who attend or accompany patients must take time off work or education. Measuring the impact of this loss-of-productivity, or economic and environmental costs is outside the scope of this project, but it is clear that if VFCs are adopted on a national scale there may be a significant knock-on effect.

Although we demonstrated a significant improvement in achieving the BOAST 7 target,⁵ we still meet this goal for only 46.4% of patients. To truly achieve this standard for all patients may require increased VFC capacity.

The high VFC discharge rate enables much-needed space to be created in face-to-face clinics for patients who need to be seen in person. The absolute reduction in patients attending face-to-face clinics has meant that clinicians potentially have more time to review each patient.

The cost-savings identified here for our local CCG represent potentially huge financial benefits nationally should VFCs be implemented more widely. However, this may present an unintended consequence whereby the Trust is financially penalised for providing this type of service, irrespective of its potential benefits in clinical parameters. Calculations from this study show that with the additional salary of the VFC nurse, the estimated loss per annum is £168 898.67 after the first year. It may be, as the BOA recommends in its recent position statement on VFCs, that: “safe provision of such services would not normally be feasible within the current tariff structure for a virtual review in the English NHS, and therefore special arrangements will normally need to be made with Trusts and commissioners”.¹⁸

Conclusions

Our findings add valuable information to the limited literature regarding VFCs. To our knowledge this is the largest and most comprehensive study addressing clinical outcomes from VFC. The results are based on robust epidemiological methodology and are significant for a number of important parameters.

Further studies are required to investigate whether similar results are reproducible across other sites, with variation in locally agreed tariffs and VFC model. Additional evaluation of the impact of VFCs on other parameters such as patient satisfaction, adverse outcomes and time-to-surgery would also be valuable.