Advancing an Algorithm for the Identification of Patients with High Data-Continuity in Electronic Health Records

Figures & data

Table 1 Attrition Table

Selection Criterion	N (Excluded)	N (Remaining)
Number of subjects in the CMS enrollment A/B/D file	–	2,435,230
At least 365 days of enrollment (A/B/D)	−374,271	2,060,959
At least 1 RPDR encounter in Medicare Part A/B/D enrollment	−1,683,796	377,163
Exclude subjects with different sex in RPDR & CMS	−275	376,888
Exclude subjects with different DOB >365 days in RPDR & CMS	−272	376,616
Age at least 65 years	−56,876	319,740
Exclude subjects with missing/incomplete data	−55,641	264,099
Final Cohort		264,099

Table 2 Parameter Estimates from Continuity Prediction Model with Demographics

Variable Name	Coefficient
Intercept	0.01
Having seen the same provider at least twice	0.03
Having general medical exam	0.10
Having BMI recorded	0.26
Having at least one diagnosis recorded	0.01
Having any ED visit	−0.02
Hospitalization(s) recorded
1	0.12
2 or more	0.12
Physician office visits recorded
1–2	0.06
3–4	0.13
5 or more	0.20
Distinct drugs recorded
1–2	0.06
3–4	0.17
5–9	0.22
10 or more	0.22
Pap smear or mammogram recorded	0.13
Colon cancer screening	0.05
Pneumococcal or influenza vaccination encounter	0.07
Age at least 74 years	−0.02
Male sex	0.02
Race
Black	0.04
Other	0.02

Notes: Parameter estimates above are from a linear regression model fit to the training data. Coefficients were selected using least absolute shrinkage and selection operator (LASSO) regression. All variables were measured in the EHR during the first year following cohort entry.

Table 3 Reclassification Table Comparing a Continuity Prediction Model with Demographics Terms versus Without Demographics Terms

Model without Demographics	Model with Demographics			Total, n (%)	Reclassified Into New Continuity Category, %
	<0.3	0.3–0.59	≥0.6		Lower	Higher	Total
<0.3
Persons included, n (%)	55,181 (76.4)	15,430 (12.4)	1595 (2.2)	72,206 (75.5)	-	14.6	14.6
High-continuity patients, n (%)	1378 (19.4)	4655 (65.6)	1060 (14.9)	7093 (28.9)	-	80.5	80.5
Low-continuity patients, n (%)	53,803 (82.6)	10,775 (16.6)	535 (0.8)	65,113 (91.6)	-	17.4	17.4
Observed proportion with high continuity, %	2.5	30.2	66.5
0.3–0.59
Persons included, n (%)	1663 (8.3)	8466 (41.1)	9963 (49.6)	20,092 (21.0)	8.3	49.6	57.9
High-continuity patients, n (%)	554 (3.9)	5380 (37.7)	8342 (58.4)	14,276 (58.2)	3.9	58.4	62.3
Low-continuity patients, n (%)	1109 (19.1)	3086 (53.1)	1621 (28.9)	5816 (8.2)	19.1	28.9	48
Observed proportion with high continuity, %	33.3	63.5	83.7
≥0.6
Persons included, n (%)	4 (0.1)	312 (9.4)	3003 (90.5)	3319 (3.5)	9.5	-	9.5
High-continuity patients, n (%)	3 (0.1)	270 (8.6)	2879 (91.3)	3152 (12.9)	8.7	-	8.7
Low-continuity patients, n (%)	1 (0.6)	42 (25.2)	124 (74.3)	167 (0.2)	25.8	-	25.8
Observed proportion with high continuity, %	75	86.5	95.9
Total
Persons included, n (%)	56,848 (59.5)	24,208 (25.3)	14,561 (15.2)	95,617 (100.0)	-	-
High-continuity patients, n (%)	1935 (7.9)	10,305 (42.0)	12,281 (50.1)	24,521 (100.0)	-	-
Low-continuity patients, n (%)	54,913 (77.2)	13,903 (19.6)	2280 (3.2)	71,096 (100.0)	-	-
Observed proportion with high continuity, %	3.4	42.6	84.3

Notes: Cells shaded in green and blue indicate patients that were re-classified into higher and lower levels of continuity when using a model that contains demographic variables, respectively.

Table 4 Difference in Measuring 40 Selected Variables in EHR vs EHR + Claims Data^a

Training Set
Year after cohort entry	1	2	3	4	5	6	7
Mean Standardized Difference (95% CI)
Top two deciles of predicted EHR-continuity	0.10 (0.08–0.13)	0.13 (0.10–0.16)	0.14 (0.11–0.17)	0.15 (0.11–0.18)	0.14 (0.11–0.17)	0.14 (0.11–0.17)	0.13 (0.10–0.16)
The remaining population	0.41 (0.30–0.51)	0.41 (0.30–0.51)	0.39 (0.30–0.48)	0.39 (0.29–0.48)	0.37 (0.28–0.46)	0.36 (0.27–0.44)	0.34 (0.26–0.42)
Validation Set
Year after cohort entry	1	2	3	4	5	6	7
Mean Standardized Difference (95% CI)
Top two deciles of predicted EHR-continuity	0.11 (0.08–0.13)	0.15 (0.12–0.19)	0.16 (0.13–0.20)	0.17 (0.13–0.21)	0.16 (0.13–0.20)	0.16 (0.13–0.20)	0.16 (0.12–0.19)
The remaining population	0.42 (0.31–0.52)	0.42 (0.32–0.53)	0.39 (0.30–0.49)	0.41 (0.31–0.51)	0.40 (0.30–0.49)	0.39 (0.30–0.49)	0.39 (0.29–0.48)

Notes: ^aLower mean standardized differences indicate less misclassification, on average, among 40 select variables when using EHR data alone vs EHR and claims data. Stable year-to-year estimates indicate consistency in the performance of the EHR-continuity algorithm.

Figure 1 Misclassification among key variable groups by decile of predicted continuity. (A) Mean sensitivity for 15 medication use variables; (B) Mean standardized difference for 15 medication use variables; (C) Mean sensitivity for 25 comorbidity variables; (D) Mean standardized difference for 25 comorbidity variables.

Figure 2 Representativeness: comparison of combined comorbidity score in those with high vs low data continuity. *Patients in the top 2 deciles of predicted care-continuity. **Patients in the remaining 8 deciles of predicted care-continuity. ***Stand diff: Standardized difference. Combined comorbidity score ranges between −2 to 26 with a higher score associated with higher mortality; smaller cell sizes were not presented here.