Referral Scoring Model: 6 Predictors, MDS Rules, Real Intake Steps

A referral scoring model turns the facts in a patient referral into a consistent risk score, so your team can decide faster and match each patient to the clinical skills on shift. We designed Smart Admissions to put this approach into daily intake.


TL;DR:

  • Weight six factors: walking limits (odds ratio 6.5), home help, patient health ratings, hospital stay, depression score, and comorbidities; treat age as an effect modifier.
  • The study’s 0.69 cut point yielded 86.3% AUC, 87.6% sensitivity, and 63.2% specificity; choose local thresholds through pilot data rather than adopting it automatically.
  • Route green referrals to prompt acceptance, yellow cases to a clinical lead, and red cases to multidisciplinary review; confirm staffing competency before offering a bed.
  • A positive PASRR Level I screen for serious mental illness or intellectual disability requires a Level II evaluation that must inform placement.
  • Validate against historical outcomes and separate samples, with clinicians reviewing every error; track readmissions, placement fit, review time, calibration, and model drift.

Smartadmissions
Turn Referral Scores Into Clearer Intake
Smart Admissions automates referral review and eligibility checks, helping post-acute teams make intake decisions with actionable insights.

Book a demo

Table of Contents

Which six predictors belong in a referral scoring system?

A peer-reviewed study of post-acute referral decisions found six significant predictors, and a separate mixed-methods study of discharge planning experts confirmed the same factors. Each one is something your admissions staff can collect from the referral packet or a short call.

The model reached an AUC of 86.3% and an overall predictive value of 83.2%, with an optimal cut-point of 0.69, a sensitivity of about 87.6%, and a specificity of about 63.2%. In plain terms, it caught most patients who needed post-acute services while flagging a fair share of patients who did not.

  • Help availability: no or intermittent help at home carries an odds ratio of about 3.0.
  • Walking restrictions: major restrictions carry about 6.5, the strongest single signal.
  • Self-rated health: good health rates about 3.1, and fair to poor rates about 4.0.
  • Length of stay: about 1.15 per additional hospital day.
  • Depression score: about 1.06 per point.
  • Comorbidities: about 1.2 per additional condition.

Age is not on this list for a reason. Experts treat age as an effect modifier, so do not give it a flat weight. Relying on age or diagnosis codes alone is also a common pitfall, because self-rated health and social support are often the better predictors and are frequently skipped in rapid workflows.

What goes into a referral scoring framework?

Think of the design in parts: inputs, scoring logic, outputs, and thresholds. Settle each before anyone writes a line of configuration.

healthcare referral scoring framework stages

For inputs, pull structured data wherever you can. The MDS 3.0 assessment items give you standardized functional and clinical fields, and the CMS RAI Manual documents item set updates you should track. Add EMR fields, payer and benefit status, documented social support, and the nursing notes that carry mobility details.

For scoring logic, you have several workable options:

  • Weighted logistic regression: transparent coefficients and easy to audit, but it needs local data to calibrate.
  • Decision-tree thresholds: simple for staff to follow, though they can miss interactions between factors.
  • Hybrid rules plus clinician override: the most practical start for most teams, because a clinician can overrule a score and you can log why.

Your outputs should include a numeric score, a green, yellow or red risk band, a suggested disposition, and the competency the receiving unit needs on shift.

Pick cut-points from your own pilot data. The study’s 0.69 cut-point favored sensitivity, which protects patients but sends more referrals to review. Raising the threshold speeds throughput and risks missing complex patients.

Pro Tip: Set your first threshold conservatively, then loosen it only after clinicians have reviewed misclassified referrals.

How do you run scoring inside the intake workflow?

Scoring is only useful if it runs where decisions already happen. Common integration points are the referral intake portal, inside the EMR, or an automated alert to your liaison. Observational work across skilled nursing facilities found three screening patterns: liaison triage, single-liaison discretion, and a multidisciplinary committee. Green, yellow and red gating fits any of them.

A workable sequence looks like this:

  1. Receive the referral and run a real-time insurance and benefit check.
  2. Complete the PASRR Level I screen. Under federal PASRR rules, a positive screen for serious mental illness or intellectual disability triggers a Level II evaluation that must inform placement.
  3. Score the referral and assign a band.
  4. Send green referrals for prompt acceptance, yellow to your liaison or clinical lead, and red to multidisciplinary review or a physician consult, by telemedicine when co-location is impractical.
  5. Check the receiving unit’s on-shift competencies against the patient’s needs before confirming the bed.
  6. Plan the MDS assessment schedule so the 5-, 14-, 30-, 60- and 90-day windows are covered from day one.

Step five matters most. The main barrier the research identifies is a mismatch between the clinical picture and real-time staffing capability. Facilities without specialized clinicians often limit intake to lower-acuity patients, and decision support helps standardize reviews to match patients to staff skills. For queue handling after scoring, see our guide to admissions queue prioritization.

How do you pilot and validate a referral evaluation model?

Test before you trust. A model that performs well in a published study still needs proof in your building, with your referral sources and your patient mix.

  1. Build a labeled dataset. Gather historical referrals with known outcomes, then calculate AUC, sensitivity and specificity for your own population.
  2. Run a controlled pilot. Keep clinicians in the loop and write success criteria in advance: 30-day readmissions, appropriateness of admissions, and review time per referral.
  3. Validate the model. Use cross-validation, have clinicians review every misclassified case, and schedule recalibration.
  4. Set governance. Get clinical leadership sign-off, train staff, document how scores are used, and define an escalation path for high-risk flags.

Research comparing algorithm advice with usual practice supports the effort. Automated decision support matches or outperforms subjective review in identifying high-risk patients and can reduce 30-day readmissions when it guides transfer decisions. A separate comparison of algorithm advice found it may improve 30-day acute care utilization.

Pro Tip: Do not tune the model to make reported readmission risk look lower. Gaming screening criteria is a known temptation, and a good score rewards matching needs to competency instead.

What should you track once the model is live?

Measure operations, clinical outcomes, and the model itself.

  • Operational: referral-to-decision time, time-to-admit, bed fill rate, and admissions staff workload.
  • Clinical: 30-day readmissions, adverse events, and appropriateness of placement.
  • Model: AUC, calibration, sensitivity, specificity, false positive and false negative rates, and drift over time.

Keep compliance on the same dashboard. Federal rules in 42 CFR Part 483 require an admissions policy, bar facilities from conditioning admission on waived rights or third-party guarantees, and call for disclosing service limitations before admission. The text of section 483.15 should shape how any score-based gate is applied.

A narrowly focused policy that accepts only low-acuity patients can harm referral volume, while a structured approach to transfer decisions can improve long-term financial outcomes by reducing avoidable readmissions. Use your metrics to confirm which side you are on. Our guide to data-driven admissions decisions goes deeper on monitoring.

What daily intake looks like with a scoring model in place

Here is our editorial view. The biggest change is not speed, though speed comes. It is the end of negotiating the same referral three different ways depending on who picked it up. A shared score gives your team a common language, and the override log shows where clinical judgment and the model disagree. Expect roles to shift: liaisons spend less time on routine referrals and more on yellow cases, and clinical leads become reviewers of edge cases rather than gatekeepers of every file.

— Harry

How Smart Admissions helps you put scoring into practice

Our platform supports healthcare referral and admissions workflows. Smart Admissions connects to your EMR and insurance portals, so eligibility checks and clinical data arrive with each referral. Your team gets:

  • Customizable decision support to help streamline referral reviews
  • Real-time eligibility verification and documentation management
  • Analytics and reporting
  • Less manual work per referral, faster review times, and faster bed fills

Onboarding is designed to be straightforward, and our support team stays with you after launch. See our subscription options on the pricing page, where the Monthly plan is $597 per month and the Annually plan is $6,447 per year.

Smartadmissions

This article is general information, not a substitute for advice from a qualified doctor. Consult a qualified healthcare professional about your own circumstances before acting on anything here.

FAQ

What is a referral scoring model in healthcare admissions?

It is a structured method that turns patient information into a numeric score or risk band for each incoming referral. Admissions teams use it to prioritize reviews and match patients to on-shift clinical capability. Decision support of this kind can standardize referral reviews.

Which factors should a referral scoring system include?

Start with help availability, walking restrictions, self-rated health, length of stay, depression score, and number of comorbidities. A mixed-methods study of discharge planning experts confirmed these six factors. Treat age as an effect modifier, not a flat weight.

Can referral scoring reduce 30-day readmissions?

It can help when scores guide transfer decisions and clinicians stay involved. Research on automated decision support found that these tools can reduce 30-day readmissions. Results depend on local validation and how consistently staff follow the output.

How do you choose a cut-point for a scoring model?

Balance sensitivity against specificity using your own pilot data. A lower cut-point catches more high-need patients but sends more referrals to review, while a higher one speeds throughput and risks missed complexity. The study’s chosen cut-point of 0.69 is a starting reference, not a universal setting.

Sources

Smartadmissions
Discuss Your Referral Scoring Workflow
Contact Smart Admissions to discuss how referral management, eligibility verification, and documentation tools can support your facility’s intake process.
Scroll to Top