DOI: 10.1111/ajag.12232
Innovation and Translation Evaluation of a New Zealand program to improve transition of care for older high risk adults Thomas E Robinson and Lifeng Zhou Waitemata District Health Board, Auckland, New Zealand
Ngaire Kerse School of Population Health, University of Auckland, Auckland, New Zealand
John DR Scott Waitemata District Health Board, Auckland, New Zealand
Jonathan P Christiansen Waitemata District Health Board; and Faculty of Medical and Health Sciences, University of Auckland, Auckland, New Zealand
Karen Holland, Delwyn E Armstrong and Dale Bramley Waitemata District Health Board, Auckland, New Zealand
Transition interventions aim to improve care and reduce hospital readmissions but evaluations of these interventions have reported inconsistent results. We report on the evaluation of an intervention implemented in Auckland, New Zealand. Participants were people over the age of 65 who had an acute medical admission and were at high risk of readmission. The intervention included an improved discharge process and nurse telephone follow-up soon after discharge. Outcomes were 28 day readmission rates and emergency attendances. The study is observational, using both interrupted times series and regression discontinuity designs. 5239 patients were treated over a one year period. There was no change in readmission rates or ED attendances or secondary outcomes. Not all patients received all components of the intervention. This transition intervention was not successful. Possible reasons for this and implications are discussed. Although non-experimental methods were used, we believe the results are robust. Key words: aged, outcome assessment (health care), patient discharge, patient readmission, telemedicine.
Objective The transition from the hospital back into the community is a crucial time for older patients. If the transition is not well supported it may lead to not only readmission, but admission to age-residential care, and poor health outcomes [1]. ‘TranCorrespondence to: Dr Thomas E Robinson, Waitemata District Health Board. Tel: +6494868920 ext. 3860; Fax: +6494418957. Email: [email protected] Correspondence to: Dr Lifeng Zhou, Waitemata District Health Board. Tel: +6494868920 ext. 2777; Fax: +6494418957. Email: [email protected] Australasian Journal on Ageing, Vol 34 No 4 December 2015, 269–274 © 2015 AJA Inc.
bs_bs_banner
sition interventions’ aim to improve the discharge process, prevent or address post discharge problems, and ensure postdischarge follow-up and support. Systematic reviews of transition interventions report a variety of types of intervention with mixed results [1–8]. We developed a programme that aimed to improve transition care for high risk older patients; including improving discharge planning by identifying high risk patients to medical and nursing teams, providing extra resources to improve discharges, and increasing post discharge support [4,6]. Individual components, such as allied health review and a discharge pharmacy service, were chosen according to clinician and manager view of needs and resources available. Telephone follow-up was attractive because of its relatively low cost, although the evidence of benefit is inconsistent [5,9]. Since transition programs are expensive, we targeted it to patients at the highest risk of readmission, using a predictive risk model [10,11]. This paper is an outcome evaluation of our programme. It aims to determine whether the intervention reduced readmissions and ED attendances in the 28 days after discharge; and secondarily look at possible reasons for the result. As a randomised controlled trial was not able to be undertaken, we also wished to investigate the feasibility of quasi-experimental methods of evaluation in this setting.
Methods Waitemata District Health Board (DHB) provides hospital services to 560 000 people in Auckland, New Zealand, including 69 000 aged 65 years and over. It has two acute general hospitals with 536 and 250 beds and provides a comprehensive range of services. Some tertiary services are provided by other Auckland hospitals. In 2011 the DHB implemented a program, called the Integrated Transition of Care (ITC) Project. The primary aim of the program was to reduce unplanned readmissions after an acute medical admission. The target population was residents aged 65 years and older, who had an acute medical admission. From this target group we identified eligible ‘High Risk’ patients using a predictive risk model (PRM) that could be run early in the patient’s admission. This type of PRM, similar to the United Kingdom’s PARR-30 tool [12], is designed to identify patients at increased risk of readmissions, although it does not predict the amenability of patients to the intervention. The model was developed by our team and uses hospital records variables including demographic information, past health service use (inpatient and outpatient), history of long term condi269
R o b i n s o n
tions (including heart failure, chronic obstructive pulmonary disease, and cancer), and medication use (polypharmacy and risk medications). The PRM had an area Harrel’s-C of 0.64 for our target population in both the development cohort and a validation cohort from a different time period. Whilst this is less than desired it is in line with international experience for similar populations [10]. The PRM was run daily and all patients who had a risk score of 20% or more were designated High Risk patients and enrolled in the ITC project. Patients identified as High Risk in testing before the ITC project started were 75% more likely to be readmitted within 28 days of discharge than low risk patients. High Risk patients were identified to clinical staff as requiring a focused discharge planning process by having a coloured bookmark inserted into their notes. The bookmark listed the three specific inpatient interventions that were requested: nutrition screening, and if necessary referral to a dietitian; allied health review; and discharge medicines reconciliation and patient education by a pharmacist. The post-discharge component was provided only to patients who were returning to independent community living, and who were not being followed by another intensive postdischarge service (e.g. palliative care services, oncology or renal services). It consisted of a telephone assessment, education, and support by a team of experienced community nurses on the first and third days post-discharge. The team was supported by a geriatrician, a pharmacist, and cultural support workers. The service aim was to identify problems not dealt with during the discharge process, assist with selfmanagement, and ensure appropriate health and social supports. Records were communicated to primary care providers electronically and via mail, and when required, by telephone. We measured whether the intervention was delivered according to the initial plan. In particular we measured whether patients received the discharge pharmacy and the nurse telephone follow-up services using program administrative records. We measured discharge process quality at the follow-up phone calls. Patients were asked if, in their view, the hospital discharge process was ‘good’, ‘satisfactory’, or ‘poor’. The telephoning nurse was asked to judge whether any important issues had been missed during the discharge process. Primary outcomes were 28 day readmission and Emergency Department attendance. We also measured readmissions over 7 and 90 days, and 28 day mortality. Loss to follow-up was very low as 98.7% of attendances of our target group were to hospitals for which we had records. We planned subgroup analyses by risk level, age group, ethnicity, and diagnostic group. For the purposes of the ITS study we defined a ‘PreIntervention Period’ (1 January 2009 to 30 November 2011) 270
T E ,
Z h o u
L ,
K e r s e
N
e t
a l .
which was prior to the intervention starting, a ‘Development Period’ (1 December 2011 to 31 March 2012) where the intervention was trialled in small numbers of patients, and an Intervention Period (1 April 2012 to 31 March 2013) where all High Risk patients were intended to be provided the new service. In our study no control group was available, as it was not possible to recreate the PRM, and therefore obtain an equivalent group, in other districts of New Zealand. Instead we use both regression discontinuity (RD) and interrupted time series (ITS) methods, which are regarded as strong quasi-experimental designs [13–15]. Both these methods propose that the outcomes are examined along a continuous ‘assignment variable’, and at the intervention ‘threshold’, which is imposed by the investigator, any difference in outcomes is likely to be due to the treatment. That is, the comparison of outcomes around this ‘discontinuity’ or ‘interruption’ will be unbiased. For RD methods the assignment variable is the variable used to determine treatment (risk score), whilst in an ITS the assignment variable is time. For the RD analysis we included patients during the Intervention Period only. We examined outcomes (28 day readmission and ED attendance) against patients’ risk scores. The expectation is that proportion of readmissions and ED attendances will increase with increasing risk scores; but if the intervention is effective there will be a drop, or discontinuity, at the risk score where the patients begin to receive the intervention (20% risk). Both parametric and non-parametric models are used to identify this discontinuity [16,17]. For the ITS we examined average readmission and ED attendance rates for High Risk patients measured monthly over the Pre-Intervention, Development, and Implementation Periods. We retrospectively calculated risk scores for patients in the Pre-Intervention period to allow us to identify patients who would have been included if the program had been available. To identify any discontinuity at the time of the start, we autoregressive integrated moving average (ARIMA) models to account for potential serial correlation of outcomes over time or seasonal effects. The internal validity of ITS studies are subject to ‘history threats’, that is, other changes occurring in the health system that may be an alternative explanation for changes in outcomes [18]. To identify the impact of history threats, we examined the outcomes in a ‘non-equivalent control group’, that is, low risk patients. We also examined two measures of patient complexity cost weight, which is derived from Diagnostic Related Group coding and other indicators of complexity, and mean length of stay (LOS) over the same time frames [13,19]. Our intervention should not have any impact on these variables, but we hypothesised that changes in many ‘history threats’, such as environment, patient mix, or health service changes, would lead Australasian Journal on Ageing, Vol 34 No 4 December 2015, 269–274 © 2015 AJA Inc.
Tr a n s i t i o n
o f
c a r e
p r o g r a m
e v a l u a t i o n
Table 1: Characteristics of included High Risk admissions Pre-intervention Demographic Female (n, %) Male (n, %) Age (years, mean, SD) European (n, %) Maori (n, %) Pacific (n, %) Asian (n, %) Complexity Risk score or predicted risk of readmission (%, median, p25-p75) Cost weight (median, p25-p75) Length of stay (days, median, p25-p75) Outcomes Readmitted within 28 days (n, %) Attending ED with 28 days (n, %) Total
6765 (48.4) 7220 (51.6) 77.6 (9.1) 10 243 (73.2) 1459 (10.4) 1758 (12.6) 525 (3.8)
Development
Intervention
P value
708 (46.4) 817 (53.6) 78.2 (9.3) 1150 (75.4) 145 (9.5) 179 (11.7) 51 (3.3)
2486 (48.1) 2686 (51.9) 78.2 (9.2) 3873 (74.9) 526 (10.2) 602 (11.6) 171 (3.3)
0.348
29 (23–38) 0.89 (0.63–1.22) 4 (2–7)
29 (24–39) 0.90 (0.58–1.22) 4 (2–7)
29 (24–39) 0.90 (0.58–1.23) 4 (2–7)
0.694 0.135 0.227
3588 (25.7) 4302 (30.8) 13 985
400 (26.2) 495 (32.5) 1525
1370 (26.5) 1721 (33.3) 5172
0.483 0.003 —
0.001 0.193
Note: All three groups of patients are used in the ITS analysis, only the Intervention patients for the RD analysis. ED, emergency department.
to changes in these variables. If we saw such changes we would be alerted to possible alternative explanations for changes in study outcomes. All analyses were undertaken in Stata 13.1 (Stata Corp LP, Texas) and SAS version 9.3 (SAS Institute, Inc., Cary, NC). The evaluation received ethical approval in 2012 (NTX/12/ EXP/077). Patient consent was not required.
Figure 1: Proportion of discharges patients identified as being either ‘satisfactory’ or ‘poor’, and proportion of discharges in which telephone follow-up nurses identified an issue having being missed, over the Intervention Period.
Results Table 1 provides information on the study patients. The average age of patients was 78 years, and most were European. About one quarter were readmitted to hospital and one third attended ED within 28 days. Patients in the three periods had similar demographic characteristics although Pre-Intervention patients were slightly younger. In the Intervention Period, 59% of eligible patients were seen by the pharmacy discharge service. Medicines reconciliations were undertaken in 79% of these patients and counselling provided to 73%. Nurse telephone follow-up was achieved for 65% of patients. Most people who were not referred to this service were living in age-residential care or had other forms of post-discharge follow-up. Of patients contacted, 68% had one call, 24% two-calls, and 8% three or more calls. Needs, such as for assistance with medication, self-management, or functional, physical or mental health issues, were identified; but only 16% of these patients required a referral to another service. Figure 1 shows that over the time of the Project there was a significant improvement in the discharge process, with significant decreases in nurse and patient identified problems over time (both P < 0.001). No data were available prior to the Intervention Period as comparison. Figure 2 shows the ITS analysis for the program, with monthly 28 day readmission rates over the study period. A Australasian Journal on Ageing, Vol 34 No 4 December 2015, 269–274 © 2015 AJA Inc.
moving average readmission rate is also shown as it reduces the ‘noise’ of the monthly plot. There is no reduction of readmissions in either the Development Period or the Intervention Period, which is confirmed by the ARIMA model (Table 2). Analyses of outcomes for the non–equivalent control group (low risk patients), and of complexity measures (WIES and LOS) also found no changes over the time period. Figure 3 shows the non-parametric RD analysis for the Intervention Period with a clear positive association between 28 day readmission rate and risk score. There is a small discontinuity at the Intervention cut-off (with an increased readmission rate) but this is not significant (P = 0.177). This analysis was not sensitive to the use of different bandwidths. We also fitted parametric linear, quadratic, and cubic models to the 271
R o b i n s o n
T E ,
Z h o u
L ,
K e r s e
N
e t
a l .
Table 2: ARIMA models for readmission rate and Emergency Department attendance rate within 28 days, High Risk patients Parameter
Estimate
Standard error
P value
Readmission Model Intercept (%) Pre-intervention period trend (% change per month) Shift (Pre-intervention/Development) (%) Development period trend (% change per month) Shift (Development/Intervention) (%) Intervention period trend (% change per month)
26.90 0.00 −1.60 0.40 1.70 −0.60
0.80 0.00 3.10 1.10 2.50 1.10
Innovation and Translation Evaluation of a New Zealand program to improve transition of care for older high risk adults Thomas E Robinson and Lifeng Zhou Waitemata District Health Board, Auckland, New Zealand
Ngaire Kerse School of Population Health, University of Auckland, Auckland, New Zealand
John DR Scott Waitemata District Health Board, Auckland, New Zealand
Jonathan P Christiansen Waitemata District Health Board; and Faculty of Medical and Health Sciences, University of Auckland, Auckland, New Zealand
Karen Holland, Delwyn E Armstrong and Dale Bramley Waitemata District Health Board, Auckland, New Zealand
Transition interventions aim to improve care and reduce hospital readmissions but evaluations of these interventions have reported inconsistent results. We report on the evaluation of an intervention implemented in Auckland, New Zealand. Participants were people over the age of 65 who had an acute medical admission and were at high risk of readmission. The intervention included an improved discharge process and nurse telephone follow-up soon after discharge. Outcomes were 28 day readmission rates and emergency attendances. The study is observational, using both interrupted times series and regression discontinuity designs. 5239 patients were treated over a one year period. There was no change in readmission rates or ED attendances or secondary outcomes. Not all patients received all components of the intervention. This transition intervention was not successful. Possible reasons for this and implications are discussed. Although non-experimental methods were used, we believe the results are robust. Key words: aged, outcome assessment (health care), patient discharge, patient readmission, telemedicine.
Objective The transition from the hospital back into the community is a crucial time for older patients. If the transition is not well supported it may lead to not only readmission, but admission to age-residential care, and poor health outcomes [1]. ‘TranCorrespondence to: Dr Thomas E Robinson, Waitemata District Health Board. Tel: +6494868920 ext. 3860; Fax: +6494418957. Email: [email protected] Correspondence to: Dr Lifeng Zhou, Waitemata District Health Board. Tel: +6494868920 ext. 2777; Fax: +6494418957. Email: [email protected] Australasian Journal on Ageing, Vol 34 No 4 December 2015, 269–274 © 2015 AJA Inc.
bs_bs_banner
sition interventions’ aim to improve the discharge process, prevent or address post discharge problems, and ensure postdischarge follow-up and support. Systematic reviews of transition interventions report a variety of types of intervention with mixed results [1–8]. We developed a programme that aimed to improve transition care for high risk older patients; including improving discharge planning by identifying high risk patients to medical and nursing teams, providing extra resources to improve discharges, and increasing post discharge support [4,6]. Individual components, such as allied health review and a discharge pharmacy service, were chosen according to clinician and manager view of needs and resources available. Telephone follow-up was attractive because of its relatively low cost, although the evidence of benefit is inconsistent [5,9]. Since transition programs are expensive, we targeted it to patients at the highest risk of readmission, using a predictive risk model [10,11]. This paper is an outcome evaluation of our programme. It aims to determine whether the intervention reduced readmissions and ED attendances in the 28 days after discharge; and secondarily look at possible reasons for the result. As a randomised controlled trial was not able to be undertaken, we also wished to investigate the feasibility of quasi-experimental methods of evaluation in this setting.
Methods Waitemata District Health Board (DHB) provides hospital services to 560 000 people in Auckland, New Zealand, including 69 000 aged 65 years and over. It has two acute general hospitals with 536 and 250 beds and provides a comprehensive range of services. Some tertiary services are provided by other Auckland hospitals. In 2011 the DHB implemented a program, called the Integrated Transition of Care (ITC) Project. The primary aim of the program was to reduce unplanned readmissions after an acute medical admission. The target population was residents aged 65 years and older, who had an acute medical admission. From this target group we identified eligible ‘High Risk’ patients using a predictive risk model (PRM) that could be run early in the patient’s admission. This type of PRM, similar to the United Kingdom’s PARR-30 tool [12], is designed to identify patients at increased risk of readmissions, although it does not predict the amenability of patients to the intervention. The model was developed by our team and uses hospital records variables including demographic information, past health service use (inpatient and outpatient), history of long term condi269
R o b i n s o n
tions (including heart failure, chronic obstructive pulmonary disease, and cancer), and medication use (polypharmacy and risk medications). The PRM had an area Harrel’s-C of 0.64 for our target population in both the development cohort and a validation cohort from a different time period. Whilst this is less than desired it is in line with international experience for similar populations [10]. The PRM was run daily and all patients who had a risk score of 20% or more were designated High Risk patients and enrolled in the ITC project. Patients identified as High Risk in testing before the ITC project started were 75% more likely to be readmitted within 28 days of discharge than low risk patients. High Risk patients were identified to clinical staff as requiring a focused discharge planning process by having a coloured bookmark inserted into their notes. The bookmark listed the three specific inpatient interventions that were requested: nutrition screening, and if necessary referral to a dietitian; allied health review; and discharge medicines reconciliation and patient education by a pharmacist. The post-discharge component was provided only to patients who were returning to independent community living, and who were not being followed by another intensive postdischarge service (e.g. palliative care services, oncology or renal services). It consisted of a telephone assessment, education, and support by a team of experienced community nurses on the first and third days post-discharge. The team was supported by a geriatrician, a pharmacist, and cultural support workers. The service aim was to identify problems not dealt with during the discharge process, assist with selfmanagement, and ensure appropriate health and social supports. Records were communicated to primary care providers electronically and via mail, and when required, by telephone. We measured whether the intervention was delivered according to the initial plan. In particular we measured whether patients received the discharge pharmacy and the nurse telephone follow-up services using program administrative records. We measured discharge process quality at the follow-up phone calls. Patients were asked if, in their view, the hospital discharge process was ‘good’, ‘satisfactory’, or ‘poor’. The telephoning nurse was asked to judge whether any important issues had been missed during the discharge process. Primary outcomes were 28 day readmission and Emergency Department attendance. We also measured readmissions over 7 and 90 days, and 28 day mortality. Loss to follow-up was very low as 98.7% of attendances of our target group were to hospitals for which we had records. We planned subgroup analyses by risk level, age group, ethnicity, and diagnostic group. For the purposes of the ITS study we defined a ‘PreIntervention Period’ (1 January 2009 to 30 November 2011) 270
T E ,
Z h o u
L ,
K e r s e
N
e t
a l .
which was prior to the intervention starting, a ‘Development Period’ (1 December 2011 to 31 March 2012) where the intervention was trialled in small numbers of patients, and an Intervention Period (1 April 2012 to 31 March 2013) where all High Risk patients were intended to be provided the new service. In our study no control group was available, as it was not possible to recreate the PRM, and therefore obtain an equivalent group, in other districts of New Zealand. Instead we use both regression discontinuity (RD) and interrupted time series (ITS) methods, which are regarded as strong quasi-experimental designs [13–15]. Both these methods propose that the outcomes are examined along a continuous ‘assignment variable’, and at the intervention ‘threshold’, which is imposed by the investigator, any difference in outcomes is likely to be due to the treatment. That is, the comparison of outcomes around this ‘discontinuity’ or ‘interruption’ will be unbiased. For RD methods the assignment variable is the variable used to determine treatment (risk score), whilst in an ITS the assignment variable is time. For the RD analysis we included patients during the Intervention Period only. We examined outcomes (28 day readmission and ED attendance) against patients’ risk scores. The expectation is that proportion of readmissions and ED attendances will increase with increasing risk scores; but if the intervention is effective there will be a drop, or discontinuity, at the risk score where the patients begin to receive the intervention (20% risk). Both parametric and non-parametric models are used to identify this discontinuity [16,17]. For the ITS we examined average readmission and ED attendance rates for High Risk patients measured monthly over the Pre-Intervention, Development, and Implementation Periods. We retrospectively calculated risk scores for patients in the Pre-Intervention period to allow us to identify patients who would have been included if the program had been available. To identify any discontinuity at the time of the start, we autoregressive integrated moving average (ARIMA) models to account for potential serial correlation of outcomes over time or seasonal effects. The internal validity of ITS studies are subject to ‘history threats’, that is, other changes occurring in the health system that may be an alternative explanation for changes in outcomes [18]. To identify the impact of history threats, we examined the outcomes in a ‘non-equivalent control group’, that is, low risk patients. We also examined two measures of patient complexity cost weight, which is derived from Diagnostic Related Group coding and other indicators of complexity, and mean length of stay (LOS) over the same time frames [13,19]. Our intervention should not have any impact on these variables, but we hypothesised that changes in many ‘history threats’, such as environment, patient mix, or health service changes, would lead Australasian Journal on Ageing, Vol 34 No 4 December 2015, 269–274 © 2015 AJA Inc.
Tr a n s i t i o n
o f
c a r e
p r o g r a m
e v a l u a t i o n
Table 1: Characteristics of included High Risk admissions Pre-intervention Demographic Female (n, %) Male (n, %) Age (years, mean, SD) European (n, %) Maori (n, %) Pacific (n, %) Asian (n, %) Complexity Risk score or predicted risk of readmission (%, median, p25-p75) Cost weight (median, p25-p75) Length of stay (days, median, p25-p75) Outcomes Readmitted within 28 days (n, %) Attending ED with 28 days (n, %) Total
6765 (48.4) 7220 (51.6) 77.6 (9.1) 10 243 (73.2) 1459 (10.4) 1758 (12.6) 525 (3.8)
Development
Intervention
P value
708 (46.4) 817 (53.6) 78.2 (9.3) 1150 (75.4) 145 (9.5) 179 (11.7) 51 (3.3)
2486 (48.1) 2686 (51.9) 78.2 (9.2) 3873 (74.9) 526 (10.2) 602 (11.6) 171 (3.3)
0.348
29 (23–38) 0.89 (0.63–1.22) 4 (2–7)
29 (24–39) 0.90 (0.58–1.22) 4 (2–7)
29 (24–39) 0.90 (0.58–1.23) 4 (2–7)
0.694 0.135 0.227
3588 (25.7) 4302 (30.8) 13 985
400 (26.2) 495 (32.5) 1525
1370 (26.5) 1721 (33.3) 5172
0.483 0.003 —
0.001 0.193
Note: All three groups of patients are used in the ITS analysis, only the Intervention patients for the RD analysis. ED, emergency department.
to changes in these variables. If we saw such changes we would be alerted to possible alternative explanations for changes in study outcomes. All analyses were undertaken in Stata 13.1 (Stata Corp LP, Texas) and SAS version 9.3 (SAS Institute, Inc., Cary, NC). The evaluation received ethical approval in 2012 (NTX/12/ EXP/077). Patient consent was not required.
Figure 1: Proportion of discharges patients identified as being either ‘satisfactory’ or ‘poor’, and proportion of discharges in which telephone follow-up nurses identified an issue having being missed, over the Intervention Period.
Results Table 1 provides information on the study patients. The average age of patients was 78 years, and most were European. About one quarter were readmitted to hospital and one third attended ED within 28 days. Patients in the three periods had similar demographic characteristics although Pre-Intervention patients were slightly younger. In the Intervention Period, 59% of eligible patients were seen by the pharmacy discharge service. Medicines reconciliations were undertaken in 79% of these patients and counselling provided to 73%. Nurse telephone follow-up was achieved for 65% of patients. Most people who were not referred to this service were living in age-residential care or had other forms of post-discharge follow-up. Of patients contacted, 68% had one call, 24% two-calls, and 8% three or more calls. Needs, such as for assistance with medication, self-management, or functional, physical or mental health issues, were identified; but only 16% of these patients required a referral to another service. Figure 1 shows that over the time of the Project there was a significant improvement in the discharge process, with significant decreases in nurse and patient identified problems over time (both P < 0.001). No data were available prior to the Intervention Period as comparison. Figure 2 shows the ITS analysis for the program, with monthly 28 day readmission rates over the study period. A Australasian Journal on Ageing, Vol 34 No 4 December 2015, 269–274 © 2015 AJA Inc.
moving average readmission rate is also shown as it reduces the ‘noise’ of the monthly plot. There is no reduction of readmissions in either the Development Period or the Intervention Period, which is confirmed by the ARIMA model (Table 2). Analyses of outcomes for the non–equivalent control group (low risk patients), and of complexity measures (WIES and LOS) also found no changes over the time period. Figure 3 shows the non-parametric RD analysis for the Intervention Period with a clear positive association between 28 day readmission rate and risk score. There is a small discontinuity at the Intervention cut-off (with an increased readmission rate) but this is not significant (P = 0.177). This analysis was not sensitive to the use of different bandwidths. We also fitted parametric linear, quadratic, and cubic models to the 271
R o b i n s o n
T E ,
Z h o u
L ,
K e r s e
N
e t
a l .
Table 2: ARIMA models for readmission rate and Emergency Department attendance rate within 28 days, High Risk patients Parameter
Estimate
Standard error
P value
Readmission Model Intercept (%) Pre-intervention period trend (% change per month) Shift (Pre-intervention/Development) (%) Development period trend (% change per month) Shift (Development/Intervention) (%) Intervention period trend (% change per month)
26.90 0.00 −1.60 0.40 1.70 −0.60
0.80 0.00 3.10 1.10 2.50 1.10
