home care operations research

Measuring Fax-to-Intake Handoff Latency in Home Care

A methods paper for measuring how long a faxed home care referral takes to reach a complete intake decision, and where it waits.

Staffing Care Home Research DeskPublished 12 min read
Measuring Fax-to-Intake Handoff Latency in Home Care exact-title research thumbnail
On this page17 sections

The operational question

A faxed referral is an arrival, not an intake. It enters a queue, waits to be noticed, is read, is checked against eligibility and service criteria, may trigger a callback, and eventually receives a disposition. Agencies describe this sequence as a single turnaround time, but the label conceals several waits with different causes and owners.

This protocol asks a narrower question: among faxed referrals received during a defined window, how much time elapses between confirmed receipt and a complete intake decision, and how much falls into identifiable wait states? The purpose is local process measurement, not a benchmark or compliance determination.

The primary endpoint is a complete decision recorded by an authorized intake professional: accept for evaluation, decline for a documented reason, or hold with a specific information request. A secondary endpoint is first substantive touch, the first documented review by intake staff. This separates acknowledgment from decision.

Why one average hides the queue

Means and medians describe the center of a distribution, not its shape. Fax queues are bimodal. A referral arriving before a shift change may be handled quickly, while one arriving minutes later may wait until the next business day. An average blends those paths and describes no actual referral.

From receipt to first touch

The first interval captures attention. It includes time a fax sits in a machine tray, an inbox, or an unassigned EHR queue. It is sensitive to staffing, shift structure, notification rules, and queue ownership. Report the median, 75th and 90th percentiles, and the share exceeding locally chosen thresholds.

From first touch to decision

The second interval captures work. It includes reading the packet, checking identifiers, confirming payer and service area, contacting the referral source, and documenting the disposition. This interval is more likely to be affected by missing pages, unclear orders, or eligibility questions. Mixing it with the first interval makes a staffing problem look like a clinical review problem, or the reverse.

A survival curve is often more informative than a summary statistic. Plot the cumulative share decided against elapsed hours, with separate curves for business hours and after-hours receipt, to show whether the process has a long tail and where it begins.

Timestamp sources and their limits

No single system records the whole handoff. The fax server records arrival, page count, station identifier, and sometimes a received image. The EHR records document upload, task creation, assignment, status changes, and the final disposition. A shared inbox may record delivery, while read receipts are unreliable and often disabled. The telephone system records call attempts and, if available, call duration.

Each timestamp answers a different question. Fax server time is the best available proxy for arrival, but it does not prove that a human saw the document. EHR upload time records ingestion, which may occur in batches hours later. A status change records a user action, not necessarily the moment of decision. If a coordinator decides by telephone and documents later, the call log may be closer to the decision than the EHR field.

Record the source system, field name, clock, and time zone for every timestamp. Check for clock drift and daylight saving time, which can distort after-hours analysis.

Defining events and the observation window

Start event

Use the earliest verifiable receipt timestamp, normally the fax server's received time. As a sensitivity analysis, use the first business hour after receipt to estimate latency when arrival is defined operationally rather than technically. If pages are split across transmissions, link them by station identifier, caller number, and patient identifiers before assigning a start time.

End event

Use the disposition timestamp together with evidence that an authorized person made the decision. A blank status field is not a decision. A note saying "called family" is not a disposition unless the agency defines it as one. The end event should be stable across reviewers, usually a structured field and reason code.

Set the observation window in advance, for example a calendar quarter, and extract all referrals that enter during that window. Follow each referral until the decision, the end of a fixed follow-up period, or administrative close. Record the extraction date and the version of every query. A referral that is still open at the end of follow-up is censored, not a zero.

Censoring, missing pages, and duplicate referrals

Right censoring is common. An unresolved referral has a lower bound on its latency but no observed endpoint. Dropping unresolved cases biases the result toward fast decisions. Keep them in the denominator, report the proportion censored, and use a method that accounts for censoring, such as a Kaplan-Meier estimate or a time-to-event model. For a fixed-window measure, report the share decided within the window alongside the share still open.

Missing pages create a different problem. Compare the fax server's page count with the pages attached in the EHR. A packet with missing pages may be placed on hold, so missingness is both a data quality issue and a cause of delay. Treat page completeness as a prespecified covariate or stratum, and do not exclude incomplete packets.

Duplicate referrals matter too. The same patient may be faxed by a hospital, a physician office, and a family member within a short period. Define a duplicate rule before looking at outcomes, using identifiers, source, service dates, and a time window. Distinguish a true duplicate from a revised referral. False merges shorten apparent latency by combining two waits, while false splits inflate the denominator.

After-hours and calendar effects

Fax machines do not observe business hours. A referral received at 6:58 p.m. may wait until morning, while one received at 8:02 a.m. may be opened within minutes. Report receipt by hour, day of week, weekend, and holiday, using the agency's actual holiday schedule. If an on-call intake process exists, separate referrals routed to on-call staff from those left in the standard queue.

Analyze time of day descriptively first. A model with hour, staffing, source, and payer can help, but many agencies have too few referrals for a complex specification. Collapse hours into blocks such as opening, midday, evening, and overnight, and report raw counts behind each estimate. Avoid modeled effects for cells with only a handful of cases.

Stratification, sample size, and power

Referral source and payer are not nuisance variables. A hospital discharge planner may send a complete packet with orders and demographics, while a community referral may arrive with a name and a phone number. Medicare Advantage and Medicaid managed care plans may impose different verification steps. Service type, such as personal care versus skilled nursing, may change the review path.

Stratify before pooling

Define the strata in the analysis plan, then report counts, censoring, and latency by stratum. Pool only when the process is plausibly similar. If a stratum has too few cases, present it descriptively and label it as unstable rather than forcing it into a model. An overall median can still be reported, but it should be accompanied by the stratum table.

Power cautions

Latency distributions are skewed, so sample size formulas based on means can mislead. Use a pilot to estimate the median, interquartile range, and censoring rate, then simulate the precision of the planned estimate. Report confidence intervals for every median and percentile. An agency with a few dozen referrals per month should expect wide intervals after stratification and should resist ranking staff on differences the data cannot resolve.

Audit sampling and data quality

An extract is only as good as the records behind it. Build an audit sample from the analysis frame, including decided, censored, duplicate, and incomplete cases. A second reviewer re-abstracts start time, end time, disposition, and stratum from the fax image, EHR note, and call log. Adjudicate disagreements and retain original values.

Test four things: whether the referral arrived when the fax server says, whether the EHR contains the same pages, whether the disposition is supported, and whether the recorded time matches the system event. Record exceptions by type rather than as one accuracy percentage. High agreement on disposition can coexist with poor agreement on timing.

Privacy, reproducibility, and the limits of the evidence

Faxed referrals contain protected health information. Use the minimum necessary data, restrict access by role, and avoid copying identifiable documents into analyst workspaces when a limited data set will answer the question. The HIPAA Privacy and Security Rules provide the privacy and safeguard framework. The NIST Privacy and Cybersecurity Frameworks help document risks, CISA's performance goals help prioritize baseline practices, and the GAO Green Book supplies internal-control language. NARA guidance is relevant to retaining the study file and audit evidence.

Reproducibility requires more than a spreadsheet. Keep a data dictionary, extraction query, extraction date, inclusion and exclusion lists, duplicate rule, censoring rule, and analysis code. Record every revision. Preserve the original extract in read-only form and document who had access. A later analyst should be able to reproduce the median and censoring count without guessing which query version was used.

What the evidence can and cannot prove

This design can show how a local fax queue behaved during a defined period, which wait states consumed time, and whether data quality supports operational decisions. It can compare periods or referral sources within one agency when definitions are stable. It cannot prove that a staffing change caused a latency change without a stronger design, establish compliance with CMS conditions of participation or Medicare coverage rules, tell an agency whether a referral should have been accepted, or evaluate an employee's competence from a time stamp.

Those limits do not make the measurement useless. They make it honest. A defensible latency study tells an agency where the queue actually waits, which is more valuable than a number implying precision the data do not support. If your team needs help turning a fax queue into a documented, auditable intake process, contact us. For related work, see referral intake data quality and after-hours callback response.

Source list

  1. CMS Home Health Conditions of Participation
  2. Medicare Home Health Coverage
  3. CMS Home Health Quality
  4. HHS HIPAA Privacy Rule
  5. HHS HIPAA Security Rule
  6. NIST Cybersecurity Framework
  7. CISA Cybersecurity Performance Goals
  8. GAO Green Book
  9. National Archives Records Management
  10. NIST Privacy Framework

Sources

<- Back to research