Building a Pediatric Airway Screening Platform: Questionnaire Architecture, Risk Scoring, and Referral Workflows

Clinical Context: Why Pediatric Airway Screening Matters

Pediatric airway disorders -- sleep-disordered breathing, obstructive sleep apnea, orofacial myofunctional disorders -- affect an estimated 10 to 15% of children. When unidentified, these conditions contribute to behavioral issues, poor academic performance, chronic mouth breathing, and orthodontic problems. Pediatric dental practices are in a good position to screen for these conditions because they see children regularly, but systematic screening has historically been paper-based and inconsistent.

The engineering challenge is taking a clinical screening workflow that works in a research setting and making it work at scale in a busy multi-location practice. That means digital questionnaires that parents actually complete, risk scoring that is clinically meaningful, referral tracking that does not lose patients between providers, and outcome data collection that feeds back into program improvement.

Why Paper Screening Does Not Scale

Paper-based screening has predictable failure modes. Completion rates are low -- parents are handed a form during a hectic check-in and either skip it or rush through it. Handwritten responses are sometimes illegible. Completed forms sit in a folder until a provider reviews them, which can take over a week. By then, the family has left and the urgency is lost. There is no systematic follow-up for children identified as at-risk, and no data collection for outcome tracking or program evaluation.

• Paper questionnaire completion rates are typically low -- around a third of parents complete them
• Completed forms often have illegible or incomplete responses
• Long delays between screening and provider review lose the clinical window
• No systematic follow-up process for at-risk children
• No data infrastructure for outcome tracking or program evaluation
• Referral coordination between providers is entirely manual

Building the Digital Screening Questionnaire

The questionnaire design was a collaboration between the clinical team, a pediatric sleep medicine physician, and our engineering team. The constraint was balancing clinical comprehensiveness with a completion time under 4 minutes on mobile. Research on mobile form abandonment shows a sharp drop-off after 4 minutes, so we had a hard ceiling on form length.

Questionnaire Structure

The questionnaire is 28 questions organized into 5 clinical domains: sleep quality and behaviors (snoring, mouth breathing during sleep, restless sleep, bedwetting, nighttime awakenings), daytime symptoms (fatigue, concentration difficulty, hyperactivity, morning headaches), feeding and swallowing (picky eating, choking, texture difficulty, prolonged mealtimes), oral habits (thumb sucking, tongue thrust, lip incompetence), and medical history (tonsil/adenoid history, allergies, asthma, family history of sleep apnea). Each question uses a response scale appropriate to the domain -- frequency scales for sleep behaviors, severity scales for daytime symptoms, binary for medical history.

The questionnaire is adaptive. Certain responses trigger follow-up questions -- for example, if a parent reports frequent snoring, additional questions about snoring volume, witnessed apneas, and sleeping position appear. This keeps the form short for children without symptoms while collecting detailed data when it matters. The adaptive logic is implemented as a simple rule engine that evaluates each response as it is submitted and appends conditional questions to the form.

Multi-Channel Delivery

Parents can complete the questionnaire through three channels: a tablet in the waiting room integrated with check-in, a pre-visit SMS link sent 48 hours before the appointment, or a QR code on printed materials. The pre-visit SMS approach accounts for the majority of completions -- parents fill it out at home on their own time, which produces more thoughtful responses than the rushed waiting-room experience. Responses save in real time, so a parent can start on their phone and finish on the tablet if needed.

• 28 questions across 5 clinical domains with adaptive follow-up logic
• Average completion time: about 3 minutes on mobile, under 3 minutes on tablet
• Multi-language support with clinically validated translations
• Pre-visit SMS delivery drives the majority of completions
• Real-time response saving with cross-device session continuity
• Digital delivery raised completion rates from around a third to over 90%

Risk Scoring Algorithm Design

The risk scoring algorithm transforms questionnaire responses into a clinical risk assessment that guides provider decision-making. We designed it in two phases: a rule-based scoring system developed with the clinical team, and a subsequent refinement using outcome data once enough volume was collected.

Rule-Based Scoring Framework

The initial framework assigns weighted points to each response based on clinical significance, calibrated against published pediatric sleep questionnaire validation studies (PSQ, SRBD scale, OSA-18). Each clinical domain produces a sub-score, and the composite risk score is a weighted combination. Sleep quality carries the highest weight (35%), followed by daytime symptoms (25%), feeding and swallowing (15%), oral habits (15%), and medical history (10%). These weights were set by the clinical advisory team based on the published literature and their clinical experience.

The composite score maps to three risk tiers. Low Risk (roughly two-thirds of screened children) receives educational materials and is rescreened at the next visit. Moderate Risk (about one in five) is flagged for provider evaluation at the current visit. High Risk (about one in ten) triggers an immediate provider alert and is prioritized for same-day assessment. The tier thresholds were set conservatively -- we would rather over-refer than miss a case, and the thresholds can be adjusted as we collect outcome data on sensitivity and specificity.

Refinement with Outcome Data

After collecting several hundred screenings with known clinical outcomes (specialist evaluation results), we used logistic regression to refine the scoring weights. The model identified some non-obvious patterns: the combination of mouth breathing during sleep plus difficulty with food textures was a stronger predictor than either symptom alone. Daytime hyperactivity combined with bedwetting showed elevated risk for moderate-to-severe OSA. Family history of sleep apnea roughly doubled the predictive power of snoring frequency as a standalone indicator.

• 18 clinical indicators weighted by evidence-based significance and refined by outcome data
• Three risk tiers with automated clinical pathway assignment
• Outcome-refined scoring achieved approximately 89% sensitivity and 94% specificity against specialist evaluation
• Domain-specific sub-scores allow providers to focus on the most concerning symptom clusters
• Scoring weights retrained periodically as outcome data volume grows

The Parent-Facing Portal Experience

Parent engagement determines whether a screening program succeeds or fails. If parents do not understand the results, do not trust the process, or find follow-up scheduling difficult, identified children do not get treated. The parent portal is the engagement surface for the entire screening and treatment journey.

Results Communication

After the questionnaire is scored, results are shared through the portal. We deliberately do not show a raw numerical score. Instead, the portal presents visual indicators (green/yellow/red) with plain-language summaries and specific next steps. For moderate and high risk results, the portal includes short educational videos (about 90 seconds each) explaining airway health concepts in accessible language. The domain-specific breakdown -- "your child scored in the moderate range for sleep-related breathing concerns but in the normal range for daytime symptoms" -- helps parents understand what is and is not flagged, which reduces anxiety.

This design was iterated based on parent feedback. An early version showed numerical scores and clinical terminology, which either caused alarm (high numbers) or dismissal (unfamiliar terms). The visual indicator approach with domain breakdown tested significantly better in post-screening surveys. Parents described feeling "informed rather than alarmed," which was the target emotional state for this communication.

Self-Scheduling and Referral Visibility

For children needing follow-up, the portal provides self-scheduling for both internal appointments and external specialist referrals. The scheduling widget integrates with each provider's system and shows real-time availability. This reduced the time from screening to follow-up appointment from about two weeks (when front-desk staff were scheduling by phone) to a couple of days. Parents can also see the status of their referrals directly in the portal, which reduces "where is my referral?" phone calls to the front desk.

• Visual risk communication with plain-language summaries and educational video content
• Self-scheduling for follow-up appointments with real-time provider availability
• Treatment progress tracking with milestone markers and outcome questionnaires
• Secure messaging between parents and the clinical team
• Document library with condition-specific educational resources
• Push notification reminders for upcoming appointments and follow-up questionnaires

Automated Referral Workflows

Pediatric airway treatment is inherently multidisciplinary. A child identified through screening might need evaluation by a pediatric ENT, a sleep study ordered by a sleep medicine physician, myofunctional therapy, orthodontic evaluation for palatal expansion, and ongoing monitoring by the pediatric dentist. Coordinating across 4 to 5 providers is the primary reason screening programs fail -- children get lost between providers.

Referral State Machine

We modeled referrals as a state machine with five states: Sent, Received, Scheduled, Completed, and Report Available. Each transition has a timeout that triggers automated escalation. If a referral has not moved to Scheduled within 5 business days, an automated reminder goes to both the parent and the specialist office. If it has not moved to Completed within 30 days, the referring dentist is alerted for manual follow-up. This automated tracking catches referrals that would otherwise stall, which in the previous manual process happened frequently.

When the dentist creates a referral, the system attaches screening results, clinical notes, and relevant imaging automatically. The specialist receives the referral electronically and can accept, view the clinical information, and schedule within the platform. This eliminates the back-and-forth of faxing records and phone-tag between offices. When the specialist completes their evaluation, they upload their report, which closes the loop and notifies the referring dentist.

• Referral creation auto-attaches screening results, clinical notes, and imaging
• Specialist portal for receiving, reviewing, and responding to referrals
• Five-state tracking with automated escalation at configurable timeout intervals
• Parent visibility into referral status via the portal
• Specialist report upload closes the loop and updates the referring provider
• Automated tracking substantially improved referral completion rates over the previous manual process

Care Coordination Dashboard

The practice's airway coordinator uses a dashboard that shows every child in the screening and treatment pipeline: children awaiting screening, children with completed screenings awaiting provider review, children with active referrals and their status, children in treatment with milestone tracking, and children due for follow-up reassessment. This single view makes it possible to manage the pipeline proactively rather than reactively chasing down missing information. Without it, the coordinator role would not scale beyond a handful of active cases.

Outcome Tracking and Clinical Analytics

A screening program is only valuable if you can measure whether identified children improve with treatment. We built outcome tracking around follow-up questionnaires administered at 30, 60, 90, and 180 days post-treatment initiation. These questionnaires use the same validated scales as the initial screening, enabling direct before-and-after comparison.

Outcome Data Model

The outcome framework tracks both subjective measures (parent-reported symptom changes on the same scales used at screening) and objective measures (clinical findings at follow-up visits). Objective measures include clinical airway assessment scores, tonsil grading (Brodsky scale), tongue mobility assessment (TRMR score), and post-treatment AHI values for children who had sleep studies. The key design decision was using identical scales for screening and follow-up so that change scores are directly comparable. This sounds obvious but it constrained the questionnaire design from the start -- we could not use a simplified screening instrument and a detailed follow-up instrument and then compare them.

• Follow-up questionnaires at 30, 60, 90, and 180 days with approximately 75% completion rate
• Consistent validated scales across screening and follow-up for direct comparison
• Provider-reported clinical outcome scoring integrated into the EHR workflow
• Treatment pathway effectiveness analysis comparing outcomes across intervention types
• Aggregate analytics dashboard for program evaluation and quality improvement

Analytics for Program Improvement

The analytics layer drives continuous improvement. Early data showed that children who received myofunctional therapy in conjunction with adenotonsillectomy had notably better outcomes than those who received surgery alone, which led the practice to update their referral protocol. The analytics also revealed that screening completion rates were significantly higher at locations using the pre-visit SMS approach versus tablet-only, which led to system-wide adoption of SMS-first delivery. These are the kinds of insights that are only possible when screening and outcome data are collected digitally in a structured format.

The data platform also makes it possible to generate outcome reports for clinical presentations and publications. Several providers in the practice have presented outcomes data at dental conferences, which creates a feedback loop of clinical engagement that sustains the program beyond its initial launch enthusiasm.

What We Learned

After running the screening program across multiple locations for several months, the technology platform did what it was supposed to: it made screening systematic rather than ad-hoc, captured data that was previously lost, and kept patients moving through a multidisciplinary care pathway. Some specific observations:

What Mattered Most

• Digital questionnaire completion rates were dramatically higher than paper -- this alone justified the platform, because you cannot screen patients who do not complete the form
• Pre-visit SMS delivery was the highest-performing channel and should be the default, not an afterthought
• The referral state machine with automated escalation was the most operationally impactful feature -- the previous manual referral process lost a substantial number of patients between providers
• Parent portal design required multiple iterations -- the first version with clinical terminology and numerical scores performed poorly in user testing
• Outcome data collection at the 180-day follow-up has lower completion rates than earlier intervals, which is an ongoing challenge for long-term outcome measurement

Design Decisions Worth Highlighting

Using the same validated scales for screening and follow-up was a constraint that paid off -- it made outcome measurement straightforward. The adaptive questionnaire logic kept the form short for most families while collecting detailed data when clinically relevant. The rule-based scoring framework, rather than a pure ML approach, was the right call for launch because it was transparent to clinicians and did not require an outcome dataset that did not yet exist. ML refinement came later once we had data to train on.

The modular architecture -- questionnaire engine, scoring engine, portal, referral system, outcome tracking -- means the platform can be extended to other screening domains (adult sleep-disordered breathing, TMJ screening) by adding new clinical content modules rather than building from scratch. Whether this modularity was worth the upfront complexity is a fair question, but for a practice that plans to expand their screening programs, it avoided redundant platform builds.