Disparities in Service and Clinical Outcomes in State-Wide Advanced Practice Physiotherapist-Led Services

This study explored variations in the primary service and clinical outcomes of a state-wide advanced practice physiotherapist-led service embedded in public medical specialist orthopaedic and neurosurgical outpatient services across Queensland, Australia. An audit of the service database over a six-year period was taken from 18 service facilities. The primary service and clinical outcomes were described. Variations in these outcomes between facilities were explored with a regression analysis adjusting for known patient- and service-related characteristics. The findings showed substantial positive impacts of the advanced practice model across all facilities, with 69.4% of patients discharged without a need for medical specialist review (primary service outcome), consistent with 68.9% of patients reporting clinically important improvements in their condition (primary clinical outcome). However, 15 facilities significantly varied from the state average for the primary service outcome (despite only three facilities varying in the primary clinical outcome). While this disparity in the primary service outcomes appears to be influenced by potentially modifiable differences in the service-related processes between facilities, these process differences only explained part of the variation. This study described the subsequent development of a new, more comprehensive set of service evaluation metrics to better inform future service planning.


Introduction
The Neurosurgical and Orthopaedic Physiotherapy Screening Clinic and Multi-disciplinary Service (N/OPSC & MDS) is an advanced practice physiotherapist-led model of care embedded in public hospital specialist orthopaedic and neurosurgical outpatient services across Queensland, Australia [1]. Patient referrals to these specialist medical outpatient services are initially triaged by either an N/OPSC & MDS advanced musculoskeletal physiotherapist (referred to herein as the "service leader") or medical specialist (varies across facilities). Referrals considered appropriate to be directed to the physiotherapistled service are usually patients with nonurgent musculoskeletal conditions (including neurosurgical patients with musculoskeletal conditions, e.g., neck and back disorders) potentially amenable to nonsurgical management. These eligible patients are referred to the N/OPSC & MDS for an initial assessment with the service leader. Depending on the initial assessment findings, a review by a medical specialist may be recommended

Audit and Data Extraction 2.2.1. Primary Outcomes
The primary service outcome was a discharge pathway, dichotomised as either discharged from the service with no specialist medical review required (Discharged) or reinstated for specialist medical review (Specialist RV). The primary clinical outcome was dichotomised as achieving (Responder) or not achieving (Non-Responder) a clinically meaningful change in the presenting condition based on an 11-point Global Rating of Change (GROC) scale, with scores between +2 to +5 reflecting a Responder and scores between −5 to +1 reflecting a Non-Responder [10].

Secondary Outcomes and Explanatory Variables
The secondary service-and patient-related outcomes and potential variation explanatory variables are listed in Table 1. Table 1. Secondary outcomes and potential explanatory variables (units/categories) included in the analysis. The condition specific, general health, and psychological measures were recorded at the initial consultation and discharge, with changes in these measures with respect to time representing the secondary patient-related clinical outcomes.

Secondary Service-Related Outcomes and Explanatory Variables
Outpatient Service (orthopaedic, neurosurgical)-indicates specialist medical outpatient service receiving the original patient referral. Triage Category (1-3)-Patient referrals are categorised as urgent (category 1), semi-urgent (category 2) or nonurgent (category 3), with the recommended timeframes for an initial outpatient consultation within 30, 90, and 365 days, respectively. Waiting Time (days)-time between specialist outpatient department receipt of initial referral and initial N/OPSC & MDS appointment. Management Duration (days)-time between initial N/OPSC & MDS appointment and discharge.
Review Appointments (absolute number)-number of patients receiving an N/OPSC & MDS review appointment. Non-attendance (yes/no)-number of patients not attending the final N/OPSC & MDS review appointment. Multidisciplinary Referrals (number)-number of patients referred to multidisciplinary treatment services (may be one or more of the services, as clinically indicated). Medical Specialist Case Discussion (yes/no)-number of patients for whom case discussion with a medical consultant was sought during the N/OPSC & MDS management period.

Secondary Service-Related Outcomes and Explanatory Variables
Quick Disabilities of the Arm, Shoulder and Hand (QDASH) (score/100)-self-reported disability for patients with shoulder/elbow/wrist/hand conditions [22] and a MCID of 10 points [23].

Data Analysis
Data was cleaned, coded, and quality-checked. All analyses were undertaken using SPSS v22 (IBM Corp, Armonk, NY, USA). Descriptive statistics were calculated to report the primary service and clinical outcomes, as well as the secondary service-and patient-related outcomes and explanatory variables (Study Aim 1). Paired t-tests and published Minimally Clinically Important Differences (MCID) ( Table 1) where available were used to further explore secondary patient-related clinical outcomes.
Hierarchical binomial logistic regression analyses were then conducted to explore variation of service and clinical outcomes between facilities (Study Aim 2). Primary service (Discharged and Specialist RV) and clinical (Responder and Non-Responder) outcomes (dependent variables) were modelled separately to assess their relationships with facilities and the potential explanatory service-related variables (independent variables), while additionally accounting for influences of the patient-related explanatory variables (Aim 2). Potential service-and patient-related variable multicollinearity issues were firstly evaluated using Pearson's (continuous normally distributed data) and Spearman's rho (non-normally or categorical data) coefficients [27] to determine if moderate (r s = 0.4-0.6) or strong (r s = 0.7-0.9) correlations [28,29] were evident between variables. Where there was risk of multicollinearity, only one variable was selected (investigator's choice based on clinical and service reasoning) to be carried forward to the final model.
In regression Model 1, the uncorrected relationship between the outcome and facility was evaluated. For both outcomes (service and clinical), the facility closest to the state average was coded as the "Referent" in SPSS. In Model 2, following the SPSS recommendations for hierarchical regression analyses [30], patient-related variables were entered, observing firstly their impact on the relationship between outcome and facility and, secondly, ensuring their potential influence on the impact of the explanatory service-related variables of interest (entered in the subsequent Model 3) were accounted for. As it was intended to determine the overall impact of the service-related variables, they were all entered together. Alpha was set at 0.05 for all statistical analyses.

Results
There were 29,319 eligible client records retrieved spanning six years (2012-2017). There were high rates of completion (>90%) of the service-related variables, as well as the patient sociodemographic measures, with minimal variations in data completeness across facilities. There was a lower completion rate for the primary (GROC, 55% completion) and secondary (completion rate range 33-80% at baseline and 24-64% at discharge) clinical outcome measures. Approximately 39% of patients were either discharged at their initial N/OPSC & MDS assessment or did not require review by the service leader and, therefore, were not eligible to complete the discharge clinical outcomes.

State-Wide N/OPSC & MDS Outcomes (Aim 1)
Across all the facilities, 69.4% of discharged patients did not require a medical specialist review (Primary service outcome), although this outcome varied across conditions, as shown in Table 2. As shown in Table 3, patients with spinal conditions represented the greatest proportion of patients managed (43.9% of cases), followed by patients with lower limb (32.7%) and upper limb (23.4%) conditions. Across all facilities, 68.9% of patients reported a clinically meaningful response to management within the N/OPSC & MDS, varying from 65.7% to 74.1% across conditions. Table 2. Service outcomes and explanatory variables presented as the number (n) and proportion (%) of patients within each category. The primary service outcome was a discharge pathway, dichotomised as either discharged from the service with no specialist medical review required (Discharged) or reinstated for specialist medical review (Specialist Review).  Table 3. Clinical outcomes and potential patient-related explanatory variables recorded at the initial service assessment and at discharge presented as means (95% confidence interval (CI)) or proportions (%), mean change between initial assessment and discharge, and the proportion of cases achieving a Minimal Clinically Important Difference (MCID).

Variation in Outcome Measures between Facilities (Aim 2)
The hierarchical binomial regression modelling findings for the primary service (Discharged) and clinical (Responder) outcomes are shown in Tables 4 and 5. The preliminary analyses showed many of the patient-related variables (Pain Severity, PSEQ, ODI, NDI, QDASH, and LEFS) to be significantly correlated (Spearman's rho 0.35-0.69, p < 0.001). To avoid multicollinearity in the multivariable model, only the Pain Severity and QOL variables were selected to be carried through to the multivariate analysis based on their relevance to all the conditions (investigator's judgement). The Outpatient Service and Condition Managed variables were also significantly related (Spearman's rho 0.51, p < 0.001), as were the Management Duration and Review Appointments variables (Spearman's rho 0.58, p < 0.001). Therefore, only the Condition Managed and Management Duration variables, respectively, were included in the multivariate analysis. The Box-Tidwell procedure [31] was performed using the variables remaining in the final models and the logit of the dependent variables (Pathway outcome and Clinical outcome). Both Age (p < 0.001) and Waiting Time (p < 0.001) demonstrated nonlinearity with the logit of the Clinical outcome (Responder/Non-responder) and were subsequently recoded to categorical variables for both regression models.

Discharge Pathway (Primary Service Outcome)
The three progressive hierarchical binomial regression models for the primary service outcome of Discharge Pathway (reference: returned to specialist outpatients waitlist) are shown in Table 4 (Clinic 1 was coded as the Referent). In Model 1, 14 facilities are seen to be significantly different to the Referent, reducing to 10 facilities in Model 2 (adjusted for patient-related variables) and increasing to 15 facilities in Model 3 (adjusted for servicerelated variables). The significant service variables in the final model included Waiting Time, Management Duration, Triage Category, Non-attendance, and Medical Specialist Input during N/OPSC management.
No outliers were evident for the primary service outcome based on the studentised residual range (SD) (−2.73 to 2.50 (0.98)) being within accepted parameters (≤−3.61, ≥3.61) based on 13 predictor variables in the final models [32]. The logistic regression model was statistically significant, χ 2 (43) = 3681, p < 0.001. The model explained 32% (Nagelkerke R 2 ) of the variance in the pathway outcome and correctly classified 76.8% of cases. One facility (Facility 16) had insufficient numbers and was excluded from the analyses for Models 2 and 3.

GROC (Primary Clinical Outcome)
The three progressive models of the hierarchical binomial regression for the primary clinical outcome of GROC (reference: nonresponse to management) are shown in Table 5 (Clinic 10 was coded as the Referent). In Model 1, 13 facilities were significantly different from the Referent, reducing to four facilities in Model 2 (adjusted for the patient-related variables) and reducing to three facilities in Model 3 (adjusting for service-related variables). The significant service variables in the final model included Management Duration, Triage Category, Non-attendance, and Medical Specialist Input.
No outliers were evident for the primary clinical outcome, according to the studentised residual (range (SD) −2.63 to 2.02 (1.04)) [32]. The logistic regression model was statistically significant, χ 2 (42) = 879, p < 0.001. The model explained 18.1% (Nagelkerke R 2 ) of the variance in the pathway outcomes and correctly classified 73.6% of cases. One facility (Facility 16) had insufficient numbers and was excluded from the analyses for Models 2 and 3. Table 4. Three hierarchical binomial regression models for the primary service outcome (Discharged) evaluating the relationship between the discharge pathway and Neurosurgical and Orthopaedic Physiotherapy Screening Clinic and Multi-disciplinary Service (N/OPSC & MDS) facility only (Model 1), adjusted for the patient-related variables (Model 2), and adjusted for the service-related variables (Model 3). Significant service variables in the final model included management duration, triage category, nonattendance, and medical specialist input. Shaded cells represent sites with significant variance from the referent facility (Facility 1). OR: adds ratio.

Discussion
Over the 2012-2017 audit period, nearly 70% of patients discharged from management within the state-wide advanced physiotherapist-led N/OPSC & MDS did not require a specialist medical consultation (Primary Service Outcome). This is substantial from a specialist waitlist resource management perspective, given also that less than five percent of patients discharged by the service present again to specialist medical services within 12 months [6]. Furthermore, 69% of patients for whom clinical outcomes were received reported a clinically meaningful improvement in their condition (GROC, Primary Clinical Outcome). In context, these are notable primary service and clinical outcomes given the generally high levels of pain severity (average pain score 58/100) and disability (conditionspecific index averaging 41-51/100) and low levels of function (PSFS 4/10) reported by patients at the initial consultation. In summary, these state-wide findings are consistent with earlier studies demonstrating the substantial impact the N/OPSC & MDS model of care delivers in managing orthopaedic and neurosurgical demands in Queensland's public hospitals [7,33]. The future challenge will be in implementing the optimal scale and mix of specialist medical and advanced physiotherapist-led services to address demands at the various public hospital facilities [1,33,34].
The most critical finding of the study, though, indicated the full impact of the N/OPSC & MDS model in managing orthopaedic and neurosurgical demands in these public hospitals may not as yet be realised. Fifteen facilities were observed to be significantly different to the referent facility in their primary service outcome of a discharge pathway (Table 3). While adjustments for patient-related characteristics initially reduced the variations (14 to 10 facilities in Model 2), adjustments for the service-related variables inflated the variations between facilities (from 10 to 15 facilities in Model 3). The significant service-related variables in this final model included the duration of the management period, the initial triage category, patient non-attendance to review appointments, and medical specialist input during the management period. Potentially, changes in service planning may address these significant service-related variables, although some may be challenging to modify, given that they may reflect differences in organisational procedures at different health services. While some patient-related variables also remained significant in the final model (Age, Gender, Condition Managed, QOL, and Pain Severity at the initial consultation), from a service planning perspective, these are not modifiable factors. The most notable observation, though, was the remaining level of uncertainty regarding other potential sources of facility variation (model estimated to just explain approximately 32% of the variance) in the discharge outcomes. This relatively low level of explained variance strongly suggests that the currently collated N/OPSC & MDS service-and patient-related evaluation metrics are not sufficient to comprehensively explain the observed variations between facilities. Instead, future service evaluations will need to capture a broader suite of patient case mix and service evaluation metrics to better explain the variation in the discharge pathway.
In contrast, the primary clinical outcome only varied significantly at three facilities compared to the referent facility when the model was adjusted for the known patient-and service-related variables. Similar to the findings exploring the service outcome variation, the significant but potentially modifiable service-related variables in the final model included waiting time, management period, the initial triage category, patient non-attendance to review appointments, and medical specialist input during the management period. Collectively, the findings suggest that addressing the variation in these significant servicerelated variables may reduce the facility variation for both the primary service and clinical outcomes. Similar to the primary service outcome model, the estimated strength of this model was modest (<20% explained variance), further indicating a broader suite of measures needs to be investigated in future service evaluations.
While some variation may be inevitable between facilities, variation potentially reflects suboptimal service provisions [35]. This, in turn, may result in inequitable clinical outcomes and the inefficient use of healthcare resources, highlighting the opportunity for further quality improvement [8]. Overall, these findings have provided information for future service evaluations (Study Aim 3), resulting in a revised set of standardised state-wide metrics. These include additional patient/demographic variables, service-related variables, and changes in the Patient-Reported Outcome Measures (PROMs) collected. Details of the new standardised dataset and performance indicators can be found in Appendix A. In summary, the collective findings of this current study, reviews of relevant epidemiological [36][37][38], PROMs [39][40][41], and chronic disease database (national and international) literature [42][43][44], together with information derived from consultations and collaborations with the N/OPSC & MDS facilities, underpinned the inclusion of the revised metrics. In particular, facilities were asked to consider their local operational processes concerning factors such as triage, medical specialist case discussions, and patient non-attendance, as these variables were observed in this study to have the strongest influence (odds ratios (OR) as shown in Table 3) on patients progressing to medical specialist consultations. This collaborative work also resulted in the addition of other service-related variables (e.g., reason for medical consultant input and multidisciplinary referral patterns, including funding sources, referrals for investigations and interventions, and the use of telehealth during an episode of care). A follow-up study is planned to examine if this new dataset can better explain facility variation in outcomes compared to those observed in this current study.
Despite such a large sample size, there were still some limitations of this study, particularly the amount of missing data for clinical outcome measures and some variations in the outcomes recorded at different facilities. The findings for the primary clinical outcome regression model may also be limited due to the use of the GROC measures to dichotomise the outcomes. However, the service uses the GROC, as it is a universal outcome across all conditions, incorporating perceived changes in patients by considering all factors, which may explain the higher proportion of patients achieving the MCID reported for the GROC compared to the condition-specific measures in Table 2. Another limitation is that there were limited service-and patient-related variables available in this study, making it challenging to derive strong association models to explain the variations in the outcomes between facilities. We anticipate that the next service evaluation study will permit stronger inferences regarding facility variation, underpinned by a wider suite of metrics driven by the findings of this current study.

Conclusions
The findings demonstrated a substantial positive impact of the advanced physiotherapistled services on overburdened public hospital specialist orthopaedic and neurosurgical outpatient services across Queensland, Australia. Potentially, this impact could be greater, given the observed disparities between facilities in discharge pathway outcomes, even following an adjustment for the differences in patient-related characteristics. While some significant service-related characteristics influencing these disparities between facilities were identified (duration of the management period, the initial triage category, patient nonattendance to review appointments, and medical specialist input during the management period) to be potentially addressed in future service planning, our findings highlighted the need for a more comprehensive collection of service-and patient-related metrics across facilities in the future. Subsequently, a new set of service evaluation metrics were described.

Informed Consent Statement:
A waiver of patient consent was approved by the Human Research Ethics Committee due to the use of de-identified, retrospective audit data.

Data Availability Statement:
The dataset from this study is not publicly available due to the data having been collated from multiple hospital health services, each with individual data custodians that require further approval for access. Please contact the first author (Maree.Raymer@health.qld.gov.au) regarding any data requests.

Acknowledgments:
The investigators would like to thank the clinical and administrative staff of the Neurosurgical and Orthopaedic Physiotherapy Screening Clinics and Multidisciplinary Service (N/OPSC & MDS) facilities across Queensland, Australia, who completed the outcome measures populating the service database.

Conflicts of Interest:
The authors declare no conflict of interest.

Appendix A. Revised State-Wide Dataset
The changes to the standardised dataset, from the results of this study, literature review, and consultation within the N/OPSC & MDS, are shown in Appendix A Table A1. New and modified measures, and those where additional response options were added, are identified. Appendix A Table A2 provides the definitions of the performance indicators against which the data is reported.      Number of patients who have achieved MCID for that specific outcome score (requires pre-and post-outcomes to be entered).
Number of eligible forms submitted where corresponding body region is checked

Global Outcome Scores-Initial and Discharge
This indicator provides the scores obtained at either initial assessment and/or discharge with respect to measures of pain, self-efficacy, and overall improvement.
Number of patients with valid score for either initial and/or discharge for global outcomes.

Number of eligible forms submitted
Region-specific outcome scores-Initial and Discharge This indicator examines the region-specific scores obtained at either initial assessment and/or discharge with respect to the body region for which they have sought treatment.
# of patients with valid score for either initial and/or discharge for their respective region-specific questionnaire.