Slide 1

Mayo Clinic logo

Matching patients across institutions without definitive patient identifiers

November 20, 2012

Jim Naessens, Stephanie Peterson, Ahmed Rahman, Matt Johnson, Diane Olson, Sue Visscher, Kyle Koenig

Slide 2

Agenda

• Purpose
• Literature Review
• Proposed Algorithms
• Results
• Summary
• Next Steps

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 3

Agenda

• Purpose

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 4

Purpose

Link 30 day readmissions and deaths for discharges between MN hospitals to:

• Determine the clinical effectiveness of current medical therapies with comprehensive "real life" databases
• Provide healthcare professionals with information that supports the highest possible quality of care.
• Linkage of data is a fundamental building block for a health information exchange.

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 5

Agenda

Literature Review

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 6

Literature Review: Record linkage

• The term "record linkage" was first used to describe the process of connecting two or more medical record documents by Halbert L. Dunn in 1946¹ in the late 1950s.

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 7

Literature Review- Record Linkage, cont.

At a high level:

• Deterministic matching
  • All or none matching - based on a single unique identification number was found to be insufficient due to missing values and typographical errors
• Probabilistic matching
  • Produces sensitivity and specificity based on the frequency and uniqueness of the data elements

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 8

Sensitivity: Test's ability to identify positive results

sensitivity = number of true positives over number of true positives + number of false negatives

= probability of a positive test given that the patient is ill

Example:
If 100 screening tests were done for strep throat, and 60 patients have strep throat, but only 45 patients were identified as having strep throat:

Truth x test	Test — strep throat=no	Test — strep throat=yes
Strep throat=no	32	8	40
Strep throat=yes	15	45 (row percent 45/(45+15) = .75	60
Total	47	53	100

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 9

Specificity: Ability of the test to identify negative results

sensitivity = number of true negatives over number of true negatives + number of false positives

= probability of a negatives test given that the patient is well

Example:
If 100 screening tests were done for strep throat, and 40 patients truly did not have strep throat, but only 32 patients were identified as not having strep throat:

Truth x test	Test — strep throat=no	Test — strep throat=yes
Strep throat=no	32 (row percent) 32/(32+8) = .80	8	40
Strep throat=yes	15	45	60
Total	47	53	100

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 10

Recent Literature

• Tromp, et all¹ determined that probabilistic matching on four variables outperformed deterministic matching
• 2009 systematic review by Silviera et al²
  • Record and field quality is a better determinant of accuracy than database size
  • More studies are needed to determine the accuracy of linkage procedures

¹ 1 Tromp, M, Ravelli, AC , Bonsel, GJ, Hasman, A, Reitsma, JB, Results form simulated data sets: probabilistic record linage outperforms deterministic record linkage. J Clin Epidemiol. 2011 May; 64(5):565-72

² Silveira DP, Artmann E., Accuracy of probabilistic record linkage applied to health databases: systematic review. Rev Saude Publica. 2009 Oct;43(5):875-82

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 11

Agenda

• Proposed Algorithms

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 12

Proposed Algorithms: REP Exploration

Rochester Epidemiology Project (REP):

• Unique research infrastructure system
• Links together medical records from SE MN Providers
• Conducts population-based descriptive, case-control, historical and prospective cohort, and cross-sectional research studies

What have we learned?

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 13

Proposed Algorithms: Approach

• We tested 4 different algorithms:
  • Algorithm 1: DOB, gender, 5-digit zip
  • Algorithm 2: DOB, gender, 9-digit zip
  • Algorithm 3: DOB, gender, 4-digit SSN (required for a match)
  • Algorithm 4: DOB, gender, 4-digit SSN when available, otherwise zip code

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 14

Proposed Algorithms: Methodology

• Data used to test approach:
  • Decision Support Database (DSD-billing) within Mayo Clinic Rochester (MCR)
  • Hospital claims data (data sent to MHA) from MCR
  • Mayo Clinic Health System (MCHS) data

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 15

Proposed Algorithms: 30-Day Readmission

Process diagram showing 30 day readmission matches going through the four algorithms described on page 13 and ending up in the three databases described on page 14.

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 16

Proposed Algorithms: 30-Day Death

Process diagram showing 30 day death matches going through three of the four algorithms described on page 13 (that is, not including algorithm 2) and ending up in the three databases described on page 14.

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 17

Agenda

Results

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 18

Results: 30-Day Readmissions- Algorithm 1 - MCR DSD

With this algorithm, we have a sensitivity of 98.4% and a specificity of 99.7%

Actual 30 Day Readmission (total %, row %, col %)	30 Day Readmission Algorithm		Total Discharges
	No readmit	Readmit
No readmit	31,386	95	31,481
	83.73%	0.25%	83.98%
	99.70%	0.30%
	99.70%	1.58%
Readmits	96	5,908	6,004
	0.26%	15.76%	16.02%
	1.60%	98.40%
	0.30%	98.42%
Total Discharges	31,482	6,003	37,485
	83.99%	16.01%	100.00%

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 19

Results: 30-Day Readmissions- Algorithm 2- MCR DSD

With this algorithm, we have a sensitivity of 97.3% and a specificity of 99.99%

Actual 30 Day Readmission (tot %, row %, col %)	30 Day Readmission Algorithm		Total Discharges
	No readmit	Readmit
No readmit	31,479	2	31,481
	83.90%	0.01%	83.98%
	99.99%	0.01%
	99.49%	0.03%
Readmits	162	5,842	6,004
	0.43%	15.58%	16.02%
	2.70%	97.30%
	0.51%	99.97%
Total Discharges	31,641	5,844	37,485
	84.41%	15.59%	100.00%

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 20

Results: 30-Day Readmissions- Algorithm 3- MCR DSD

With this algorithm, we have a sensitivity of 92.94% and a specificity of 100%

Actual 30 Day Readmission (tot %, row %, col %)	30 Day Readmission Algorithm		Total Discharges
	No readmit	Readmit
No readmit	31,481	0	31,481
	83.98%	0.00%	83.98%
	100.00%	0.00%
	98.67%	0.00%
Readmits	424	5,580	6,004
	1.13%	14.89%	16.02%
	7.06%	92.94%
	1.33%	100.00%
Total Discharges	31,905	5,580	37,485
	85.11%	14.89%	100.00%

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 21

Results: 30-Day Readmissions- Algorithm 4 - MCR DSD

With this algorithm, we have a sensitivity of 99.83% and a specificity of almost 100%

Actual 30 Day Readmission (tot %, row %, col %)	30 Day Readmission Algorithm		Total Discharges
	No readmit	Readmit
No readmit	31,480	1	31,481
	83.98%	0.00%	83.98%
	100.00%	0.00%
	99.97%	0.02%
Readmits	10	5,994	6,004
	0.03%	15.99%	16.02%
	0.17%	99.83%
	0.03%	100.00%
Total Discharges	31,490	5,995	37,485
	84.01%	15.99%	100.00%

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 22

Results: 30-Day Readmissions for MCR: DSD

		Sensitivity	# visits with a true readmit missed	Specificity	# falsely identified visits with a readmissions
Algorithm 1	DOB, gender, 5-digit zipcode	98.4%	96	99.7%	95
Algorithm 2	DOB, gender, 9-digit zipcode	97.3%	162	99.9%	2
Algorithm 3	DOB, gender, last 4 of SSN	92.9%	424	100%	0
Algorithm 4	DOB, gender, last 4 of SSN or 9-digit zip if unavailable	99.8%	10	100%	1

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 23

Results: 30-Day Readmissions for MCR: Claims

		Sensitivity	# visits with a readmissions algorithm missed	Specificity	# falsely identified visits with a readmissions
Algorithm 1	DOB, gender, 5-digit zipcode	97.9%	108	99.7%	81
Algorithm 2	DOB, gender, 9-digit zipcode	96.5%	174	100%	0
Algorithm 3	DOB, gender, last 4 of SSN	92.8%	358	100%	0
Algorithm 4	DOB, gender, last 4 of SSN or 9-digit zip if unavailable	99.8%	12	100%	0

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 24

Results: 30-Day Readmissions for MCR: MCHS

		Sensitivity	# visits with a readmissions algorithm missed	Specificity	# falsely identified visits with a readmissions
Algorithm 1	DOB, gender, 5-digit zipcode	97.99%	159	99.7%	169
Algorithm 2	DOB, gender, 9-digit zipcode	84.9%	1174	99.9%	29
Algorithm 3	DOB, gender, last 4 of SSN	93.5%	505	100%	0
Algorithm 4	DOB, gender, last 4 of SSN or 9-digit zip if unavailable	99.2%	64	99.97%	14

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 25

Results: 30-Day Deaths for MCR DSD

		Sensitivity	# visits with a death algorithm missed	Specificity	# falsely identified visits with a death
Algorithm 1	DOB, gender, 5-digit zipcode	85.6%	118	99.96%	15
Algorithm 2	DOB, gender, 5-digit zipcode	Cannot be applied because MN Death tapes do not provide 9-digit SSN
Algorithm 3	DOB, gender, last 4 of SSN	94.1%	49	100%	0
Algorithm 4	DOB, gender, last 4 of SSN or 5-digit zip if unavailable	95.4%	38	99.99%	4

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 26

Results: 30-Day Deaths for MCR Claims

		Sensitivity	# visits with a death algorithm missed	Specificity	# falsely identified visits with a death
Algorithm 1	DOB, gender, 5-digit zipcode	84.9%	118	99.96%	14
Algorithm 2	DOB, gender, 9-digit zipcode	Cannot be applied because MN Death tapes do not provide 9-digit SSN
Algorithm 3	DOB, gender, last 4 of SSN	93.39%	52	100%	0
Algorithm 4	DOB, gender, last 4 of SSN or 9-digit zip if unavailable	94.5%	41	99.99%	4

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 27

Results: 30-Day Deaths for MCHS Data

		Sensitivity	# visits with a death algorithm missed	Specificity	# falsely identified visits with a death
Algorithm 1	DOB, gender, 5-digit zipcode	84.4%	223	99.95%	25
Algorithm 2	DOB, gender, 9-digit zipcode	Cannot be applied because MN Death tapes do not provide 9-digit SSN
Algorithm 3	DOB, gender, last 4 of SSN	94.5%	79	99.99%	4
Algorithm 4	DOB, gender, last 4 of SSN or 9-digit zip if unavailable	95.8%	60	99.98%	9

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 28

Agenda

Summary

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 29

Summary

• We presented 4 algorithms using 3 different data sources for post-hospitalization 30-day readmission and 30-day death
• DOB, gender, 4-digit SSN along with zip code appeared to improve the overall sensitivity and/or specificity
  • Proved true for all datasets
  • Both readmissions and death data

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 30

Agenda

Next Steps

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 31

Next Steps

• Will be sending early next week

Center for the SCIENCE of HEALTH CARE DELIVERY logo

Slide 32

Questions and Discussion

Center for the SCIENCE of HEALTH CARE DELIVERY logo