When flexibility outpaces safety: rethinking pregnancy screening in random-start ovarian stimulation Authors: Randy Morris Published: 2026 The case reported in this issue, severe ovarian hyperstimulation syndrome (OHSS) precipitated by gonadotropin stimulation in a patient with an unrecognized natural conception, is presented, appropriately, as a cautionary case report (1). I would argue it is something more: an early symptom of a structural gap that the field has opened for itself and has not yet closed. Random-start ovarian stimulation (RSOS) protocols represent a genuine clinical advance, offering scheduling flexibility that benefits both patients and clinics (1). But their adoption has quietly narrowed a margin of safety that was rarely tested before. Early-follicular-phase stimulation, usually on cycle days 2–3, required a defined menstrual event before treatment began, and was typically paired with a baseline blood test and ultrasound—a combination that, together with the bleed itself, made an unrecognized pregnancy or an inadvertent luteal-phase start extremely unlikely. Gonadotropin-releasing hormone-agonist downregulation, though also sometimes initiated in the luteal phase, offered a related protection by a different route: when pregnancy occurred during the downregulation interval; the typical result was failure to achieve suppression, not silent progression to stimulation. The pregnancy effectively announced itself by blocking the next step of the protocol. Severe OHSS from an unrecognized pregnancy was, as a result, a rare and almost paradoxical outcome under the old paradigm. Random-start ovarian stimulation removes this incidental safeguard. What remains, in most protocols, is a single baseline human chorionic gonadotropin (hCG) measurement—promoted, without anyone quite deciding so, from a backup check within a layered system to the sole safeguard in a flattened one. This pattern is well recognized in other high-reliability fields. In aviation safety, it is described as the “Swiss cheese model" of accident causation, in which multiple imperfect, independent defensive layers are stacked so that a hazard passing through a gap in one is caught by the next; accidents occur when the gaps in every layer align at once (2). In anesthesiology, similarly, layered defenses—preoperative checklists, independent double-checks of drug labeling, and capnography monitoring—are deliberately kept redundant rather than consolidated into a single check, precisely because any one layer can fail silently. The RSOS collapses reproductive medicine's equivalent layered defense into a single measurement, without anyone having evaluated whether that measurement is adequate to prevent the resulting harm. The discussion accompanying this case report proposes a sound clinical heuristic in response: monitor progesterone trends across stimulation, and pursue reflex hCG testing when levels are elevated or rising, rather than relying on trigger-day values in isolation. I do not dispute the clinical logic. Patient-safety science offers a useful lens for evaluating this recommendation. Hazard controls typically are ranked by how dependent they are on a human noticing something and acting correctly in real time. Structural safeguards (so-called forcing functions) and automation are considered the strongest methods for preventing harm; training, awareness, and vigilance are considered weak, precisely because they are the controls most vulnerable to workload, incomplete information, and competing demands in a busy clinic. Watching progesterone trends more closely is a vigilance-based intervention, and therefore is inherently weak. That this is a weak control is not hypothetical. Undetected pregnancies in luteal phase starts are not rare (3). The current case report and the two cases of inadvertent-pregnancy OHSS reported by Semrl et al. (4) all show, retrospectively, that the signal was present and interpretable before OHSS developed, and in all three, it was not acted upon in time. Three independent instances converging on the same missed signal is not evidence that clinicians need to be more careful. It is evidence that detecting this signal is a poor fit for human vigilance under real-world conditions, and that we should considering stronger methods. One possible approach: an electronic medical record–embedded solution in which any stimulation flowsheet identified as RSOS would trigger scheduled progesterone and hCG checks, with gonadotropin dosing or trigger held pending results, aiming to halt an inadvertent cycle as early as possible, and ideally before trigger, rather than relying on a clinician to catch an unexpected trend in real time. Any such rule, however, runs into a problem the current literature has not yet confronted directly. Using serial progesterone levels to infer an occult pregnancy requires knowing what trajectory to expect for a given patient. Oligo-anovulatory patients with polyendocrine metabolic ovarian syndrome, patients with hypothalamic amenorrhea, and patients undergoing luteal- or late-follicular-phase starts are, by definition, those for whom cycle day and standard ultrasound findings are poor surrogates for endocrine phase, a point this case illustrates directly. A single universal progesterone cutoff, whatever value the literature eventually converges on, will not generalize cleanly across these subgroups, because their baseline progesterone physiology differs. The reported values, a median of 12.12 ng/mL in undetected pregnancies in one series, vs. the markedly different trajectories in the Semrl et al. cases (4) and in the present case, hint at this heterogeneity already. What is needed is not a single number, but population-stratified or phase-agnostic thresholds, validated prospectively rather than inferred case by case. The procedure of RSOS was not conceived as a method to reduce monitoring. Its purpose was to decouple stimulation start from cycle day, not to cut visits or laboratories. But as private equity ownership of in vitro fertilization practices expands, monitoring visits are a direct and easily trimmed cost center, and a protocol adopted for flexibility creates a convenient opening for monitoring to shrink for reasons of margin rather than medicine. Fewer scheduled visits can be marketed to patients as convenience, even when the actual driver is reduced cost per cycle rather than a clinical judgment that the visit was unnecessary. The result is an even thinner safety net, marketed as convenience. Finally, it is worth acknowledging what our evidence actually consists of. The data reported so far, including in this case, all come from retrospective case series and back-calculation. That approach is how the problem was first recognized, and case reports like this one remain valuable for that reason. However, retrospective cases alone cannot build a validated screening tool. What the field needs now is prospective, multicenter testing of a combined threshold-and-trend algorithm, with formal measures of sensitivity and specificity, rather than continued reliance on plausibility argued after the fact. The recommendation that follows from this case, then, is not simply “watch progesterone more closely.” It is that professional societies, such as the American Society for Reproductive Medicine and the European Society of Human Reproduction and Embryology, should move toward a standardized, ideally automatable, decision rule for pregnancy screening in RSOS protocols, one built on prospectively validated, population-stratified thresholds rather than retrospective heuristics applied by individual clinicians under time pressure. Random-start ovarian stimulation has changed how flexibly we can treat patients. It is time the safety architecture around it changed to match. Declaration of Interests R.M. has nothing to disclose. References