Table 2. Effect of the Intervention on Occurrence and Quality of Patient-Clinician Communication About Advance-Care Planning^a.

Outcome	Patients/Clinicians, No.b	Target Visit, Percentage or Mean (95% CI)c		β (95% CI)d	P Value
		Control	Intervention
Patient Report of Occurrence of a Goals-of-Care Discussion at Clinic Visit, %
All patientse,f,g	395/121	31 (25 to 38)	74 (68 to 81)	1.25 (0.94 to 1.56)	<.001
Patients who did not object to discussione,f,h	303/115	28 (21 to 35)	78 (72 to 85)	1.47 (1.13 to 1.81)	<.001
EHR Documentation of Goals-of-Care Discussion at Clinic Visit, %
All patientse,i	492/123	17 (12 to 22)	62 (55 to 68)	1.25 (0.92 to 1.58)	<.001
Patients who did not object to discussione,i	379/121	17 (12 to 22)	63 (56 to 70)	1.29 (0.92 to 1.66)	<.001
QOC Score at Target Visit, Mean Rating
Overall QOC about end-of-life issues
4-Indicator construct—items most supported by the interventionj	268/108	2.13 (1.00 to 3.25)	4.59 (1.75 to 7.42)	2.02 (0.48 to 3.57)	.01
Individual QOC itemsk
1. Talking about patient’s feelings about getting sickerl	338/116	6.30 (5.60 to 6.99)	7.70 (7.06 to 8.37)	2.209 (0.934 to 3.48)	.001
2. Talking about details of getting sickerm	338/117	5.95 (5.25 to 6.62)	6.96 (6.25 to 7.66)	1.24 (−0.33 to 2.80)	.12
3. Talking about how long patient might have to liven	345/118	3.43 (2.76 to 4.10)	4.03 (3.28 to 4.77)	2.34 (−0.426 to 5.08)	.10
4. Talking about what dying might be likeo	353/119	2.10 (1.53 to 2.66)	2.23 (1.60 to 2.86)	1.57 (−1.74 to 4.89)	.35
5. Discussing patient’s wishes for end-of-life treatmentl	340/115	4.50 (3.77 to 5.23)	6.72 (5.97 to 7.48)	4.63 (2.06 to 7.19)	<.001
6. Asking about things in life that are important to patientl	351/120	5.64 (4.92 to 6.36)	7.15 (6.47 to 7.82)	2.40 (0.90 to 3.91)	.002
7. Asking about patient’s spiritual or religious beliefsp	347/118	2.43 (1.82 to 3.03)	3.11 (2.44 to 3.88)	2.46 (−0.25 to 5.16)	.08

Abbreviations: EHR, electronic health record; CFA, confirmatory factor analysis; QOC, quality of communication; WLSMV, weighted least squares with mean and variance adjustment.

^aThe coefficient estimates and P values were based on complex single-group regression models with patients clustered by treating clinician, with the intervention group as the predictor of interest, and with adjustment for any variables that confounded the association between the intervention-group predictor and outcome. A variable was identified as a confounder if its addition to the bivariate model changed the coefficient for the intervention group by 10% or more in either direction. For all outcomes, patients were included only if they provided data for the outcome and any covariates in the model.

^bNumber of patients/number of clinician clusters.

^cFor occurrence of discussion, the unadjusted percentage is the percentage of each group with a qualifying discussion; for the individual QOC items, the unadjusted mean is the mean outcome at the target visit. For the multi-indicator construct, the mean of the construct on the after-visit questionnaire, based on a 2-group CFA model using patients with complete data for all 4 indicators at baseline and after visit, with scalar measurement invariance imposed between groups and over time, and with the baseline mean for the control group fixed at 0.

^dPoint estimate and confidence interval for regression coefficient from single-group model with outcome regressed on the intervention-group predictor, adjusted for any confounders.

^eBinary outcome modeled with probit regression, estimated with WLSMV.

^fAdjusted for occurrence of discussion prior to study enrollment.

^gA supplementary analysis included 369 patients, excluding patients in the intervention group whose clinicians did not report whether they had used the Jumpstart-Tips form. Compared with patients in the control group, and adjusting for occurrence of discussion prior to enrollment, patients whose clinicians did not use the Jumpstart-Tips form (n = 20) were somewhat, although not significantly, more likely to report a discussion (β = 0.57, P = .07), whereas those whose clinicians used the Jumpstart-Tips form (n = 138) were significantly more likely to report a discussion (β = 1.46, P < .001).

^hA supplementary analysis included 280 patients, excluding patients in the intervention group whose clinicians did not report whether they had used the Jumpstart-Tips form. Compared with patients in the control group, and adjusting for occurrence of discussion prior to enrollment, both patients whose clinicians did not use the Jumpstart-Tips form (n = 12, β = 0.90, P = .02) and those whose clinicians used the Jumpstart-Tips form (n = 108, β = 1.65, P < .001) were significantly more likely to report a discussion at the target visit.

ⁱNo confounders; bivariate model. (No adjustment was made for documentation of discussions prior to the patient’s study enrollment because chart abstractions did not extend back to the preenrollment period.)

^jBased on cases with complete data for all 4 indicators at both time points. Ordered categorical indicators modeled with probit regression and estimated with weighted least squares with mean and variance adjustment; adjusted for the baseline level on the construct.

^kOutcomes (0-11 ratings) defined as censored from below and modeled with Tobit regression, estimated with WLSMV.

^lAdjusted for baseline value on the outcome variable.

^mAdjusted for baseline value on the outcome variable and clinician specialty.

ⁿAdjusted for baseline value on the outcome variable, clinician type, and clinician specialty.

^oAdjusted for baseline value on the outcome variable, patient racial/ethnic minority status, education, and income; clinician type and specialty.

^pAdjusted for baseline value on the outcome variable and clinician type.