What the dashboard can miss

Research-backed clinical review priorities beyond app engagement: medical safety, hidden behaviors, screening limits, practical barriers, and digital harms.

A complete food log, reassuring score, or quiet concern column does not establish recovery or safety. Use this guide alongside the patient's treatment plan to decide what needs a conversation, medical assessment, or change in support. It complements Stepping down Pippa support.

1. Medical deterioration at any body size

Evidence: Garber et al. (2019), studying 116 hospitalized young people with anorexia or atypical anorexia, found that weight-loss history contributed information about illness severity beyond admission weight. This was an analysis within an inpatient trial, not proof that an app-derived weight trend can triage someone safely. Paper

Ask and act: Review recent intake, rate and duration of weight change, dizziness/fainting, weakness, hydration, and relevant physical symptoms. Arrange medical assessment and appropriate observations or investigations through the treating medical provider; do not wait for a low BMI or a dashboard flag. The SAHM position paper (2022) addresses medical management of restrictive eating disorders in adolescents and young adults, including atypical anorexia and ARFID. Its age-specific recommendations should not be applied uncritically to every adult.

Acute symptoms such as collapse, chest pain, confusion, or inability to maintain fluids warrant urgent assessment through the local clinical pathway. Refeeding risk and the level of monitoring needed are medical decisions; meal photographs and app estimates cannot resolve them.

2. Purging and medication-related compensation

Ask and act: Ask privately about vomiting, laxatives/diuretics, water loading, fasting, and medication use or omission, without assuming motive. A logged meal may coexist with compensation. NICE NG69 recommends assessment of fluid/electrolyte balance when compensatory behaviors are suspected and consideration of ECG monitoring based on relevant risks. Use medical assessment rather than a reassuring food log.

Evidence for a specific subgroup: Goebel-Fabbri et al. (2008) followed women with type 1 diabetes for 11 years; insulin restriction was associated with increased morbidity and mortality. The observational, women-only cohort does not supply a risk estimate for every patient, but supports asking about missed insulin and coordinating promptly with diabetes care. Paper

Do not give medication-adjustment instructions through this guide. If omission is reported, clarify access, fear of hypoglycemia, and other reasons as well as weight-related intent. The purge follow-up is a conversation aid, not medical clearance.

3. Exercise that becomes compulsory

Evidence: Monell et al. (2018) examined over 9,000 patients in a clinical database. Continuing or starting compulsive exercise during treatment was associated with poorer remission outcomes; baseline compulsive exercise alone did not predict remission. This is observational evidence, not proof that exercise caused the outcome. Paper

Ask and act: “What happens if you cannot exercise?” Ask about guilt, compensating for meals, rigid rules, secret activity, and activity despite illness or injury. Revisit this during treatment, including when food logs improve. Step counts or minutes do not establish compulsivity. Agree an individual activity plan with the treatment team rather than prescribing a universal exercise ban.

4. Self-harm and suicidal distress despite reassuring scores

Evidence: A 2023 systematic review and meta-analysis examined self-injury and suicidal thoughts/behaviors in eating disorders relative to healthy and psychiatric controls. Group associations support clinical attention; they cannot predict an individual patient's next action.

Ask and act: Ask directly about current distress, self-harm, suicidal thoughts, immediate safety, and available support when clinically indicated. Follow the clinic's assessment and crisis pathway. Do not wait for a positive automated alert, and do not treat acknowledgment of an alert as resolution.

NICE NG225 advises against using scores or global risk categories to predict suicide or decide access to care or discharge. A Pippa concern band organizes information; it is not a suicide prediction. Review safety planning and the crisis response workflow with a named clinician responsible for follow-up.

5. Symptoms outside the questionnaire's coverage

Evidence: Kutz et al.'s SCOFF diagnostic-accuracy review (2020) found important limits in the evidence across eating-disorder presentations and populations. Screening performance in selected samples does not establish reliable exclusion in every clinic. Burton Murray et al.'s NIAS validation study (2021) examined ARFID-related sensory, appetite, and fear presentations; it also highlights the need to consider other eating-disorder symptoms when interpreting an ARFID screen.

Ask and act: Explore sensory aversion, fear of choking/vomiting, low interest in eating, loss-of-control eating, and weight/shape or muscularity concerns as appropriate. Consider medical contributors to restricted eating. Select an assessment suited to the presentation, language, and patient rather than treating a negative SCOFF or improving total score as a diagnosis or remission. NIAS is discussed as research here; this does not imply it is a built-in Pippa measure.

6. Food access and practical barriers mistaken for avoidance

Evidence: An eight-year longitudinal study (2022) linked adolescent household food insecurity with later disordered eating. Longitudinal association supports checking material circumstances, but does not establish the reason for one patient's missed meal.

Ask and act: “Were the foods in the plan affordable and available this week?” Also ask about cooking/storage facilities, transport, work schedules, and a private place to eat. Adapt the plan and connect appropriate practical support. Separately check device access, connectivity, language, and accessibility before interpreting missing logs as refusal or loss of motivation. Those digital checks are practical suggestions, not findings from the food-insecurity study.

7. Tracking that increases pressure or reduces honesty

Evidence: Lindgreen et al. (2018) observed 41 patients and interviewed 26 using Recovery Record. Features and clinician visibility could be experienced as supportive or obstructive, including a sense of surveillance. The qualitative design identifies experiences, not their prevalence in Pippa. Paper

Ask and act: “Is there anything you avoid recording because of how it might look to us?” Ask about repeated photo retakes, correcting logs until they feel right, delaying meals to record them, and distress around rewards. These are proposed prompts, not validated markers. Agree the minimum useful recording for the treatment plan and review burdensome features together. Clarify who can see which information, response times, and the limits of monitoring. Support honest disclosure without rewarding perfect records. See app checking and stepping down support.

8. AI advice becoming an authority or replacing care

Evidence: The 2023 paper “Ethical Challenges in AI Approaches to Eating Disorders” discusses risks including inappropriate advice. It is an ethics discussion, not a trial measuring how often Pippa causes harm. Evidence for one chatbot cannot validate another model, version, or use.

Ask and act: “Has any AI answer changed how you eat, move, take medication, or seek help?” Ask about advice conflicting with the care plan and repeated requests for permission or reassurance. Review the relevant exchange with appropriate consent; clarify inaccurate advice and assess its impact. Report the problem through the clinic's approved support route without putting patient content in a public issue. Agree an alternative if a feature is harmful. Do not simply tell the patient to ask the same chatbot again, and do not promise that filters catch every unsafe response.

Use these priorities in routine care

At onboarding, establish the medical and mental-health assessment arrangements, practical access needs, and who reviews app information. During planned reviews, revisit relevant domains and anything that has changed. A brief dashboard scan cannot replace this clinical work.

These are workflow suggestions, not validated urgency categories. Use the treating team's judgment and local protocols. For each concern, document the patient's account, the evidence available and missing, the action, the responsible person, and the follow-up date. Confirm that referrals or contacts happened. A note or email alone does not complete the response.

Evidence reviewed September 7, 2026. This page adds review priorities; it does not add automated detection, diagnostic capability, or continuous clinical monitoring to Pippa.