palliative connect: triggered palliative care …...palliative care already consulted 6 (10.3) no...

1
Palliative Connect: Triggered Palliative Care Consultation Using an EHR Prediction Model Katherine R. Courtright, MD, MS 1,2,3 , Corey Chivers, PhD 4 , Michael Becker, BS 4 , Susan H. Regli, PhD 4 , Linnea Pepper, MD 3 , Robert L. Stetson, MHA 5 , Michael Draugelis, BS 4 , Nina O’Connor, MD, FAAHPM 1,3 o Frequency and timing of palliative care consultation are highly variable among patients with different life-limiting illnesses o Triggering palliative care consultation based on certain diagnoses in the EHR is increasing, but this strategy is a poor surrogate for actual needs and is neither equitable nor sustainable RATIONALE 1 Palliative and Advanced Illness Research Center, Perelman School of Medicine; 2 Center for Health Incentives and Behavioral Economics; 3 Department of Medicine, Perelman School of Medicine; 4 Predictive Analytics, University of Pennsylvania Health System; 5 Corporate Office of Strategic Decision Support, University of Pennsylvania Health Systems, all at the University of Pennsylvania o Evaluate the feasibility and clinical impact of triggering palliative care consultation based on predicted risk of 6-month mortality OBJECTIVE Prediction model development o Model development phase: 70/30 split of admissions (N=64,246) to 3 hospitals at 1 academic center in 2016; multivariate logistic regression model including demographics, comorbidities POA, laboratory values, admission type; C-statistic and calibration curve o Trigger evaluation phase: pre-post pilot study on hospitalist services at 1 urban, academic hospital; Intervention : triggered (with opt-out) palliative care consult on hospital day 2 if risk of 6- month mortality ≥0.3 (IRR 100%); Outcomes : palliative care processes, clinical outcomes, and direct costs METHODS RESULTS Figure 1. EHR 6-month mortality prediction model AUROC CONCLUSIONS Figure 2. EHR 6-month mortality prediction model calibration Consult trigger evaluation Table 1. Characteristics of patients in the pre-post intervention cohorts Characteristic* Control ( n =142) Intervention ( n =134) Patient age (year), median (IQR) 72.5 (65.5, 81.6) 72.6 (63.0, 83.0) Female, n (%) 54 (38.0) 58 (43.3) Race, n (%) White 83 (58.5) 69 (51.5) Black 52 (36.6) 59 (44.0) Asian 5 (3.5) 3 (3.3) Other/Unknown 2 (1.4) 3 (2.2) Married, n (%) 83 (58.5) 72 (53.7) Admission type urgent, n (%) 142 (100) 133 (99) Elixhauser Index, median (IQR) 9 (6, 12) 8 (6, 12) Palliative Connect score, mean (SD) 0.5 (0.2) 0.5 (0.2) Abbreviations: IQR, interquartile range; SD, standard deviation *p<0.05 for all comparisons between control and intervention cohorts Reasons for declined triggered consults ( n =58) n (%) No palliative care needs at this time 24 (41.4) Primary team meeting palliative care needs 8 (13.8) Discharge anticipated soon 8 (13.8) Hospice already consulted 6 (10.3) Palliative care already consulted 6 (10.3) No reason provided 4 (6.9) Other 2 (3.5) Table 2. Reasons from primary team for declining triggered palliative care consultation Table 3. Intention-to-treat analysis of triggered palliative care consultation among patients with predicted 6-month mortality risk ≥0.3 Measure Control ( n =142) Intervention ( n =134) p-value Clinical outcomes In-hospital mortality, n (%) 7 (5.0) 2 (1.5) 0.11 Hospital length of stay (day), median (IQR) 5.7 (3.5, 9.8) 5.9 (3.9, 10.6) 0.50 ICU admission, n (%) 33 (23.2) 19 (14.2) 0.05 ICU length of stay (day), median (IQR) 4.4 (1.4,6.2) 2.7 (1.9,4.7) 0.50 30-day all-cause readmission* 29/130 (22.3) 23/127 (18.1) 0.40 Palliative care processes Palliative care consult order 23 (16.2) 84 (62.7) <0.001 Pre-consult length of stay (day), median (IQR) 2.8 (1.3, 5.8) 1.2 (0.7,2.7) 0.0013 Advance care planning documentation, n (%) 24 (16.9) 36 (24.9) 0.05 Change in code status, n (%) 43 (30.3) 40 (29.9) .094 Outpatient palliative care referral, n (%) 6 (4.3) 22 (16.4) <0.001 Hospice discharge*, n (%) 13 (9.2) 23 (17.2) 0.05 Economic outcomes Total hospital direct costs, median (IQR) $8,814 (5,623, 20,070) $9,088 (5,365, 16,648) 0.92 o An EHR risk stratification tool reliably identifies patients with high risk of mortality within 6 months who would not otherwise have received palliative care consultation o Triggering inpatient palliative care consultation based on predicted 6- month mortality risk is feasible; leads to earlier and more frequent high quality palliative care; and may improve clinical outcomes LIMITATIONS o A single center study at a large academic center with a multidisciplinary palliative care team o Non-randomized study design o Several (n=14) eligible patients unable to be offered a triggered consult due to palliative care team strain FUTURE DIRECTIONS o Expanded intervention across all medical services at 2 hospitals (ongoing) o Interviews with stakeholders to explore perceptions of and preferences for EHR triggers (ongoing) o Determine optimal strategy for palliative care delivery at different risk thresholds This study was funded in part by a career development award from the National Palliative Care Research Center (KRC) Author contact: [email protected]

Upload: others

Post on 22-Aug-2020

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Palliative Connect: Triggered Palliative Care …...Palliative care already consulted 6 (10.3) No reason provided 4 (6.9) Other 2 (3.5) Table 2. Reasons from primary team for declining

Palliative Connect: Triggered Palliative Care Consultation Using an EHR Prediction ModelKatherine R. Courtright, MD, MS1,2,3, Corey Chivers, PhD4, Michael Becker, BS4, Susan H. Regli, PhD4, Linnea Pepper, MD3, Robert L. Stetson, MHA5, Michael Draugelis, BS4, Nina O’Connor, MD, FAAHPM1,3

o Frequency and timing of palliative care

consultation are highly variable among patients with different life-limiting

illnesses

o Triggering palliative care consultation based on certain diagnoses in the EHR is

increasing, but this strategy is a poor surrogate for actual needs and is neither

equitable nor sustainable

RATIONALE

1Palliative and Advanced Illness Research Center, Perelman School of Medicine; 2Center for Health Incentives and Behavioral Economics; 3Department of Medicine, Perelman School of Medicine; 4Predictive Analytics, University of

Pennsylvania Health System; 5Corporate Office of Strategic Decision Support, University of Pennsylvania Health Systems, all at the University of Pennsylvania

o Evaluate the feasibility and clinical

impact of triggering palliative care consultation based on predicted risk of

6-month mortality

OBJECTIVE

Prediction model development

o Model development phase: 70/30 split

of admissions (N=64,246) to 3 hospitals

at 1 academic center in 2016;

multivariate logistic regression model

including demographics, comorbidities

POA, laboratory values, admission type;

C-statistic and calibration curve

o Trigger evaluation phase: pre-post pilot

study on hospitalist services at 1 urban,

academic hospital; Intervention:

triggered (with opt-out) palliative care

consult on hospital day 2 if risk of 6-

month mortality ≥0.3 (IRR 100%);

Outcomes: palliative care processes,

clinical outcomes, and direct costs

METHODS

RESULTS

Figure 1. EHR 6-month mortality prediction model AUROC

CONCLUSIONS

Figure 2. EHR 6-month mortality prediction model calibration

Consult trigger evaluationTable 1. Characteristics of patients in the pre-post intervention cohorts

Characteristic* Control (n=142) Intervention (n=134) Patient age (year), median (IQR) 72.5 (65.5, 81.6) 72.6 (63.0, 83.0) Female, n (%) 54 (38.0) 58 (43.3) Race, n (%) White 83 (58.5) 69 (51.5) Black 52 (36.6) 59 (44.0) Asian 5 (3.5) 3 (3.3) Other/Unknown 2 (1.4) 3 (2.2) Married, n (%) 83 (58.5) 72 (53.7) Admission type urgent, n (%) 142 (100) 133 (99) Elixhauser Index, median (IQR) 9 (6, 12) 8 (6, 12) Palliative Connect score, mean (SD) 0.5 (0.2) 0.5 (0.2)

Abbreviations: IQR, interquartile range; SD, standard deviation *p<0.05 for all comparisons between control and intervention cohorts

Reasons for declined triggered consults (n=58) n (%) No palliative care needs at this time 24 (41.4) Primary team meeting palliative care needs 8 (13.8) Discharge anticipated soon 8 (13.8) Hospice already consulted 6 (10.3) Palliative care already consulted 6 (10.3) No reason provided 4 (6.9) Other 2 (3.5)

Table 2. Reasons from primary team for declining triggered palliative care consultation

Table 3. Intention-to-treat analysis of triggered palliative care consultation among patients with predicted 6-month mortality risk ≥0.3

Measure Control (n=142) Intervention (n=134) p-value Clinical outcomes In-hospital mortality, n (%) 7 (5.0) 2 (1.5) 0.11 Hospital length of stay (day), median (IQR) 5.7 (3.5, 9.8) 5.9 (3.9, 10.6) 0.50 ICU admission, n (%) 33 (23.2) 19 (14.2) 0.05 ICU length of stay (day), median (IQR) 4.4 (1.4,6.2) 2.7 (1.9,4.7) 0.50 30-day all-cause readmission* 29/130 (22.3) 23/127 (18.1) 0.40 Palliative care processes Palliative care consult order 23 (16.2) 84 (62.7) <0.001 Pre-consult length of stay (day), median (IQR) 2.8 (1.3, 5.8) 1.2 (0.7,2.7) 0.0013 Advance care planning documentation, n (%) 24 (16.9) 36 (24.9) 0.05 Change in code status, n (%) 43 (30.3) 40 (29.9) .094 Outpatient palliative care referral, n (%) 6 (4.3) 22 (16.4) <0.001 Hospice discharge*, n (%) 13 (9.2) 23 (17.2) 0.05 Economic outcomes Total hospital direct costs, median (IQR) $8,814 (5,623, 20,070) $9,088 (5,365, 16,648) 0.92

o An EHR risk stratification tool reliably identifies patients with high risk of

mortality within 6 months who would

not otherwise have received palliative

care consultation

o Triggering inpatient palliative care consultation based on predicted 6-

month mortality risk is feasible; leads to

earlier and more frequent high quality

palliative care; and may improve clinical outcomes

LIMITATIONS

o A single center study at a large academic

center with a multidisciplinary palliative

care team

o Non-randomized study design

o Several (n=14) eligible patients unable to be offered a triggered consult due to

palliative care team strain

FUTURE DIRECTIONS

o Expanded intervention across all medical

services at 2 hospitals (ongoing)

o Interviews with stakeholders to explore

perceptions of and preferences for EHR

triggers (ongoing)

o Determine optimal strategy for palliative

care delivery at different risk thresholds

This study was funded in part by a career development award from the National Palliative Care Research Center (KRC)Author contact: [email protected]