Abstract
Validation Study of a Clinical Decision Support Algorithm for Hypothyroidism Management During Pregnancy Original title (Greek): Μελέτη επικύρωσης αλγορίθμου υποστήριξης κλινικών αποφάσεων για τη διαχείριση υποθυρεοειδισμού κατά την κύηση 1.2 Authors Dr. Konstantinos Toulis, MD (Principal Investigator, algorithm developer) Co-investigators: Dr Stylianos Mandanas 1.3 Background and Rationale The management of clinical and subclinical thyroid disorders during pregnancy represents a significant clinical challenge, affecting 2–5% of pregnant women in Greece. While the American Thyroid Association (ATA) and European Thyroid Association (ETA) guidelines provide general recommendations, their clinical application shows considerable variability among physicians due to the flexibility of therapeutic options. Clinical decision support systems (CDSS) have proven effective in improving care quality and reducing clinical variability. In the field of pregnancy endocrinology, no validated decision support system currently exists for managing the assessment of treatment necessity, provision of individualised levothyroxine dosing advice, and dose adjustment based on treatment response. This study aims to validate a CDSS algorithm for hypothyroidism management during pregnancy by evaluating the clinical appropriateness of its recommendations against the judgement of specialist endocrinologists with at least 10 years of clinical experience. 1.4 Hypotheses Primary Hypothesis (Non-Inferiority): The clinical decision support algorithm achieves a clinical appropriateness rate (Tier 1 + Tier 2 + Tier 3 concordance) of ≥95%, with the lower bound of the 95% confidence interval remaining above the non-inferiority margin of 85%. Secondary Hypotheses: (a) The rate of clinically inappropriate recommendations (rated “Clinically Inappropriate” or “Potentially Harmful” by expert assessors) is less than 5% of total cases. (b) The proportion of patients achieving trimester-specific TSH targets at first follow-up (4–8 weeks) does not differ based on the tier of algorithm-expert concordance. 2. Study Design 2.1 Design Prospective validation study with consecutive recruitment of pregnant women presenting for thyroid function evaluation. The study follows standard clinical care; the algorithm generates recommendations in parallel but does not influence treatment decisions. Patients receive standard care from the investigator-clinician independently of the study. 2.2 Group Sequential Design The study employs a group sequential (adaptive) design, permitting interim analyses and potential early stopping for either efficacy or futility. O’Brien-Fleming stopping boundaries are used, which are conservative early in recruitment, requiring strong evidence for early stopping and thereby controlling the overall Type I error rate. Planned analyses: Analysis Sample Size Stopping Rule 1st Interim Analysis n = 24 Stop for efficacy if observed appropriateness exceeds O’Brien-Fleming boundary; stop for futility if probability of demonstrating non-inferiority at final analysis is negligible 2nd Interim Analysis n = 48 Same criteria with updated boundaries Final Analysis n = 72 Definitive non-inferiority assessment Recruitment may be terminated early if non-inferiority is established or futility is demonstrated at an interim analysis. Maximum recruitment duration is 12 months. 2.3 Blinding Expert assessors complete Sections A and B of the Expert Assessment Form (primary recommendation and acceptable alternatives) before the algorithm’s recommendation is revealed. Only after recording their independent judgement do assessors view the algorithm output and complete Sections C and D (clinical appropriateness rating and discordance justification). The algorithm generates recommendations automatically from structured clinical input data without knowledge of physician decisions. 2.4 Study Procedures Patient Enrolment (T0) — Standard Clinical Care: The investigator-clinician collects clinical and laboratory data in real time and provides individualised treatment recommendations as part of routine care, independently of the study. Parallel Validation Procedures: Eligibility screening per inclusion/exclusion criteria. Consecutive enrolment of all eligible patients. Real-time entry of clinical and laboratory data into the CDSS. Recording of the expert endocrinologist’s primary recommendation and one explicitly stated alternative (established by consensus of two independent expert assessments). Recording of automated algorithmic recommendations. Documentation of agreement/disagreement using the structured Expert Assessment Form. Follow-Up (T0 + 4–8 weeks) — Standard Clinical Care: Repeat clinical and laboratory evaluation as part of routine care. TSH measurement, treatment compliance assessment, patient satisfaction. 2.5 Clinical Input Variables Variable Description TSH (mIU/L) Most recent thyroid stimulating hormone measurement Gestational age (weeks) Current week of pregnancy at assessment (4–40 weeks) Trimester T1: ≤13 weeks; T2/T3: >13 weeks TAI status Anti-TPO and/or anti-TG: positive, negative, or unknown Body weight (kg) Categorised: 75 kg Treatment status De novo (no prior LT4) or on-treatment (current dose) Current LT4 dose (mcg) If on treatment: current daily levothyroxine dose Clinical history Thyroidectomy, recurrent hypothyroidism Risk factors Twin pregnancy, IVF conception, history of first-trimester miscarriage Absorption concerns Intestinal surgery, inflammatory bowel disease Concurrent medications Iron supplements, calcium supplements Cardiac history Arrhythmias, cardiovascular disease (exclusion criterion for some; flag for others) 3. Sampling Plan 3.1 Existing Data Registration prior to analysis: This registration is submitted while the study is actively recruiting. As of the registration date, 52 participants have been enrolled out of a planned maximum of 72. No data have been analysed. No interim or final analyses have been conducted. The protocol-specified interim analyses at n=24 and n=48 were deferred due to operational considerations and will be conducted per protocol concurrently with the final analysis. No study data have been examined for hypothesis testing purposes. All analyses will follow this pre-registered plan. 3.2 Inclusion Criteria Pregnant women aged 18–45 years; gestational age 4–40 weeks (including multiple pregnancies); referral for thyroid assessment or finding of thyroid dysfunction; required baseline data available (TSH, gestational age, weight, medical history, concurrent medications); signed informed consent. 3.3 Exclusion Criteria Hyperthyroidism or thyrotoxicosis; history of thyroid cancer or thyroid hormone resistance syndrome; severe comorbidities with increased morbidity (chronic renal failure, hepatic failure); arrhythmias (atrial fibrillation, tachyarrhythmias, atrioventricular block); epilepsy; psychiatric disorders preventing cooperation; participation in another clinical study; inability to attend follow-up. 3.4 Sample Size Maximum sample: N = 72 participants. Sample size rationale: Based on the group sequential design with the following parameters: Parameter Value Primary endpoint Proportion of clinically appropriate recommendations Expected appropriateness (p₁) 95% Non-inferiority margin (p₀) 85% Statistical power (1–β) 80% Significance level (α) 0.05 (one-sided) Interim analyses 2 interim + 1 final Stopping boundaries O’Brien-Fleming The study may terminate before n=72 if an interim analysis demonstrates efficacy or futility. 3.5 Recruitment Setting: Endocrinology, Diabetes and Metabolism Outpatient Clinic, 424 General Military Hospital of Education (424 GSNE), Thessaloniki, Greece. Method: Consecutive enrolment of all eligible patients presenting for thyroid evaluation during pregnancy. Maximum recruitment period: 12 months. Current status: 52 of 72 participants enrolled.