HI499 Unit 10 Capstone Final Paper Example

From 4.09 Percent Back to 4.09 Percent: A Unit 10 Capstone Final Paper Arguing for One Decision on Duplicate Records From Online Self-Scheduling

[Student Name]

Purdue University Global

HI499: Bachelor's Capstone in Health Information Management

Unit 10 Capstone Final Paper

[Instructor Name]

[Date]

Callowfield Health, its people and every figure are composite; the research, guidance and rules cited are real.

Executive Summary

Of every hundred patient records Callowfield Health's online self-scheduling tool creates, about four turn out to belong to someone who already had a record. The system as a whole runs at 1.79 percent. The cause is a setting chosen when the tool launched: when the matching engine finds a likely but uncertain match, the tool creates a new record instead of asking a person. This paper recommends routing those uncertain matches to a staff review queue, standardizing addresses before matching and adding two questions to the booking screen. The plan takes sixteen weeks, costs $16,800 once and saves a net $5,443.83 a year, and its main purpose is safer records rather than savings. The decision requested is approval of Phase 1 at the next steering meeting.

The Problem and Its Measure

The project measures one thing: of the records the tool creates in a given month, the proportion a specialist later confirms already existed under another record, with each month's count closed 60 days after it ends so that late discoveries are included. The baseline is 4.09 percent, the average of twelve closed months. The target is 1.0 percent, reached within half a year of go-live and then held for three closed months in a row. Emergency registrations, clinic front desks and the old backlog of duplicate pairs are outside the project, each for a stated reason: different causes, an already low rate and no effect on new duplicates.

Local Data

Twelve months of the merge log held 2,544 confirmed pairs. Set against each channel's volume of new records, the figures single out the tool: it accounts for 18.2 percent of new records and 41.5 percent of confirmed duplicates. Clinic desks run at 0.66 percent, the call center at 1.49 and emergency registration at 1.83. The tool's rate did not drift down across the year, so the problem was not a launch issue fading on its own, and rising volume meant more duplicates each month. Most were found by accident: 61 percent surfaced only when the patient came back and someone noticed a missing history, typically 19 days later, and care often took place on the wrong record in that window. Duplicates from the tool held about 23 claims a month at billing and took roughly 66 hours of specialist time to merge, and two of the system's three overlays in the year involved a record the tool had created.

What the Evidence Says

Published research and professional reviews agree on one point that bears directly on the tool: matching engines produce a middle band of uncertain scores, and those cases belong with a trained reviewer rather than an automatic rule in either direction (Riplinger et al., 2020). Studies of index data show that the fields behind most mismatches are names, birth dates and addresses, entered differently for ordinary reasons (Just et al., 2016), and an evaluation on real records found that standardizing such fields improves matching accuracy (Grannis et al., 2019). Research on identifiers has tied split records to safety risks such as missing allergy information (McCoy et al., 2013). The evidence is thinner on patient-entered data, and no published study was found that measures duplicates from self-scheduling tools, so Callowfield's own baseline is presented as local evidence, not a general rate. The field's advice and the tool's setting point in opposite directions: every source sends uncertain matches to a person, and the tool sent them to record creation.

The Root Cause

When 120 of the tool's confirmed pairs were drawn at random, they showed what differed between each pair's two records: name variants in 46, new contact details in 29, an address change or format difference in 21, a mistyped birth date in 14 and other differences in 10. Most were honest differences, not careless typing. The engine's logs then showed that 97 of the 120 newer records had matched an existing record in the uncertain band. The tool had found the patient and created a new record anyway, because the launch configuration chose record creation over a held booking to protect booking completion, with no risk review and no policy behind the choice. Patient carelessness and staff error were considered and rejected as root causes.

Constraints and the Redesign They Forced

Four fixes were tested against privacy, breach, record amendment, state identifier and claims rules before any cost was counted. The staff review queue survived, open only to the data integrity role so that reviewers see no more than a match decision needs (45 C.F.R. § 164.502). Address standardization survived under a business associate agreement. Showing patients a list of possible matching records was blocked, because it would disclose other people's names and birth dates, and it became two questions that reveal nothing, one about earlier visits and one about any former surname. A required full Social Security number was narrowed to an optional last four digits under California law and internal minimization policy. One correction came from seminar: a classmate wanted to know how the design treats a mother or father booking on behalf of a son or daughter. The review had never asked who was typing. The redesign now flags proxy bookings for a relationship check, never tells the person booking whether an existing record was found and labels the last-four field as the patient's own.

Implementation and Cost

The plan runs in four four-week phases: configuration and testing, address standardization, queue staffing and training, then go-live and a 90-day watch. The address service goes live before the queue opens, so reviewers do not spend time on pairs it would have separated. One-time costs are $14,500 for the vendor's configuration change and $2,300 for training. Recurring costs are $8,400 a year for the address service and $9,062.40 for queue review time, 2,832 reviews at six minutes each. Savings, counted at the 1.0 percent target rather than zero, are $19,152.00 in merge labor and $3,754.23 in claim rework, leaving a net of $5,443.83 a year and a payback of about three years. The queue adds about 24 hours of review a month but removes about 50 hours of merging, so no positions are added.

Evaluation

The outcome measure is the tool's duplicate rate under the proposal's definition, read monthly once each month closes. Two balancing measures guard against harm from the change: the abandonment rate for online bookings, which triggers an alert if it climbs two points past its 11 percent baseline, and queue age, with 95 percent of held bookings resolved within one working day of arrival. A process measure tracks the share of tool records with a standardized address. Each reading is plotted over time, and only a lasting shift, never a single good month, is accepted as change (Langley et al., 2009). Because the counting window is 60 days, the first post-change month cannot be read until two months after go-live, and the target cannot be confirmed until at least five months after it. The final reading goes to the steering committee carrying the same definition footnote as the baseline.

Limits

Confirmed duplicates undercount the true number, although the undercount should be similar before and after the change. The sample of 120 pairs is large enough to show a dominant cause but not to rank the smaller ones precisely. Cost figures rest on stated assumptions about minutes and wages that a finance reviewer may adjust. And the plan's strongest benefit, fewer overlays, is too rare to measure within a year; two or three events a year cannot show a trend quickly, so the paper relies on the logic that fewer duplicates mean fewer merges and fewer chances for a wrong one.

Conclusion

The capstone began with a merge log almost nobody read and one figure in it: 4.09 percent. The data showed that the figure was local and stable, the evidence showed what the field recommends, the cause analysis found the setting that contradicted it, the rules reshaped the fix and the plan showed it can be done for a modest cost. Approving Phase 1 is the decision that lets the next report show a different number, and the evaluation is built so that whoever reads that report will know whether the 4.09 percent has really fallen.

References

45 C.F.R. § 164.502 (2024). https://www.ecfr.gov/current/title-45/section-164.502

Grannis, S. J., Xu, H., Vest, J. R., Kasthurirathne, S., Bo, N., Moscovitch, B., Torkzadeh, R., & Rising, J. (2019). Evaluating the effect of data standardization and validation on patient matching accuracy. Journal of the American Medical Informatics Association, 26(5), 447-456. https://doi.org/10.1093/jamia/ocy191

Just, B. H., Marc, D., Munns, M., & Sandefer, R. (2016). Why patient matching is a challenge: Research on master patient index (MPI) data discrepancies in key identifying fields. Perspectives in Health Information Management, 13(Spring), 1e.

Langley, G. J., Moen, R. D., Nolan, K. M., Nolan, T. W., Norman, C. L., & Provost, L. P. (2009). The improvement guide: A practical approach to enhancing organizational performance (2nd ed.). Jossey-Bass.

McCoy, A. B., Wright, A., Kahn, M. G., Shapiro, J. S., Bernstam, E. V., & Sittig, D. F. (2013). Matching identifiers in electronic health records: Implications for duplicate records and patient safety. BMJ Quality & Safety, 22(3), 219-224. https://doi.org/10.1136/bmjqs-2012-001419

Riplinger, L., Piera-Jiménez, J., & Dooling, J. P. (2020). Patient identification techniques: Approaches, implications, and findings. Yearbook of Medical Informatics, 29(1), 81-86. https://doi.org/10.1055/s-0040-1701984