Every year, millions of people are trained to perform cardiopulmonary resuscitation on rubbery manikins that flash scores, percentages, and encouraging messages on nearby tablets. Those numbers carry real weight: a high score can certify a nurse, a medical student, or a bystander as competent, while a low score sends the learner back to practice. But what if the score itself is quietly misleading? A new experimental study published in the Journal of Medical Systems suggests that the widely used QCPR scoring system, developed by the resuscitation training company Laerdal Medical, can award high marks to chest compressions that fall well outside international guideline recommendations, and that its output shifts depending on something most trainers never think about: the stiffness of the manikin’s chest spring.
The research team, led by Špela Metličar of the Medical Dispatch Centre Maribor at University Clinical Centre Ljubljana together with collaborators in Denmark, Slovenia, and the United Kingdom, set out to probe a surprisingly basic question. Guidelines from the International Liaison Committee on Resuscitation and the European Resuscitation Council call for compressions 5 to 6 centimeters deep at a rate of 100 to 120 per minute, with full chest recoil and minimal interruptions. The QCPR system condenses these parameters, along with hand position, recoil, and compression fraction, into a single percentage score from 0 to 100. Yet the exact mathematical formula behind that score is proprietary and unpublished, which means educators and researchers cannot independently verify how faithfully the composite number tracks guideline compliance. The team decided to test it directly, using a machine instead of a human rescuer to eliminate the usual noise of human performance.
The experimental setup was elegantly simple. A brand-new Resusci Anne QCPR manikin was placed on the floor and connected to a Corpuls mechanical chest compression device, which holds its piston in a fixed position over the compression point. That standardization matters: by using a machine, the researchers locked in perfect hand position, complete recoil, and an uninterrupted compression fraction, isolating depth and rate as the only variables. They then systematically swept through every combination of compression depth from 2.0 to 6.0 centimeters, in half-centimeter steps, and compression rates from 80 to 120 per minute, in increments of five. Each two-minute compression epoch was repeated with three interchangeable chest springs rated at 30, 50, and 60 kilograms, simulating chests of different stiffness. In total, 243 individual measurements were recorded through the SkillReporter for Tablet application.
The results reveal a scoring system dominated almost entirely by depth. At compressions of just 2.0 to 2.5 centimeters, QCPR scores were a flat zero percent regardless of rate or spring. Scores crept up at 3.0 centimeters, ranging from 10 to 18 percent, and climbed further at 3.5 centimeters, reaching 31 to 51 percent. The most volatile region was 4.0 to 4.5 centimeters, where scores swung dramatically with both rate and spring resistance; at 4.5 centimeters, scores ranged from 49 to 99 percent, with most exceeding 80. From 5.0 centimeters onward, scores were consistently high, between 57 and 100 percent, and became nearly independent of compression rate. In other words, the algorithm rewards pushing deep and largely stops caring about much else once you do.
The spring resistance findings are perhaps the most unsettling. Even with identical preset depth and rate settings, swapping the chest spring changed QCPR scores by up to 18 percent, observed at 4.5 centimeters of depth and rates of 80 and 85 compressions per minute. Differences of 8 to 11 percent appeared at other depths, and even at 5.5 and 6.0 centimeters, where scores approached the maximum, spring-related gaps of up to 4 percent persisted. The mechanism remains uncertain. Because the scoring algorithm is closed, the authors cannot determine whether different spring stiffnesses alter the compression waveform, the recoil dynamics, mechanical oscillation, or the response of the sensors themselves, or whether force-sensitive inputs feed into the score. What is clear is that a difference of this magnitude can flip a learner across a competency threshold, meaning two trainees delivering mechanically identical compressions could receive different verdicts depending on which manikin chest they happened to practice on.
To quantify how well the score maps onto actual guidelines, the team analyzed agreement between the manufacturer’s threshold of 75 percent, the cutoff for an Advanced CPR Performer, and truly guideline-compliant compression epochs, defined as 5 to 6 centimeters deep at 100 to 120 per minute. Of the 243 epochs, only 45, or 18.5 percent, met the guideline definition. The 75 percent QCPR threshold caught every single guideline-compliant epoch, a sensitivity of 100 percent, but its specificity was just 70.2 percent and its positive predictive value a sobering 43.3 percent. Cohen’s kappa, a measure of agreement beyond chance, came in at 0.47, indicating only moderate agreement. Sensitivity analyses at thresholds of 70 and 80 percent told the same story. Put plainly, more than half of the epochs earning a passing grade were delivered at depths or rates outside the recommended ranges.
The implications ripple outward from the simulation lab. In a classroom, a trainee who compresses too shallowly but at a favorable rate might see a respectable score and assume their technique is sound, receiving positive reinforcement precisely when correction is needed. An instructor relying on the composite number might miss the specific parameter that is failing. In certification contexts, a learner could be deemed ready to perform guideline-compliant CPR when their actual compressions would fall short in a real cardiac arrest. The authors are careful to note that manikin findings cannot be extrapolated directly to patient outcomes, but the educational stakes are hard to overstate: feedback systems exist precisely to shape behavior, and a score that diverges from the target behavior risks shaping the wrong one.
There is also a clinical dimension the study could only partially explore. The mechanical device’s operational limits capped testing at 6.0 centimeters, so the researchers could not determine whether the QCPR system penalizes excessively deep compressions at all. This matters because a major registry-based study published in JAMA Cardiology found the highest survival at compression depths of 4.5 to 5.0 centimeters and rates of 100 to 110 per minute, with outcomes declining on both sides of that sweet spot. If the scoring algorithm does not punish depth beyond 6 centimeters, it may overemphasize deep compressions and tolerate broader rate ranges than clinical evidence supports, potentially nudging trainees toward the very compressions associated with injury risk in patients.
The authors’ prescription is refreshingly straightforward. Rather than layering more adjustments onto an opaque composite score, future feedback systems should prioritize transparent, guideline-based reporting: the percentage of compressions with guideline-compliant depth, guideline-compliant rate, complete recoil, and overall compliance. This aligns with the Resuscitation Education Utstein consensus recommendations for uniform reporting in resuscitation research. Some researchers have proposed open-access, AI-driven scoring models trained on large real-world datasets as an alternative to closed proprietary algorithms, and the authors see room for such innovations, but only as complements to, never replacements for, guideline-anchored metrics. Until scoring systems are adjusted for or proven independent of manikin chest stiffness, they argue, spring resistance should always be reported, and QCPR scores should be read alongside absolute depth and rate values.
The study has limits worth keeping in view. Each depth, rate, and spring combination was tested only once, though the mechanical compressor minimized variability, and the preset device settings were not independently verified by the scoring system itself. The researchers did not measure compression force or the full waveform, so they cannot say whether score differences arose from compression dynamics, sensor behavior, or the hidden algorithm. Only continuous compressions were tested, a single manikin was used, and unit-specific calibration effects cannot be excluded, although the manikin was new and had received less than 10 percent of the roughly 500,000 compressions that trigger manufacturer-recommended component replacement. The complete dataset and R analysis code are publicly available on Zenodo, an openness that itself models the transparency the authors are calling for. For the millions of trainees who will face a glowing tablet this year, the message is clear: a high CPR score is a signal, not a guarantee, and the numbers beneath the number still matter most.
Subject of Research: Validation of the QCPR chest compression feedback scoring system against guideline-compliant CPR parameters in a manikin model
Article Title: Association of Chest Compression Depth, Rate, and Spring Resistance with “Quality CardioPulmonary Resuscitation” (QCPR) Scores in a Manikin Model
Article References: Metličar, Š., Lauridsen, K. G., Štiglic, G., & Fijačko, N. (2026). Association of Chest Compression Depth, Rate, and Spring Resistance with “Quality CardioPulmonary Resuscitation” (QCPR) Scores in a Manikin Model. Journal of Medical Systems, 50(1), Article 142. https://doi.org/10.1007/s10916-026-02469-z
Image Credits: AI Generated
DOI: 10.1007/s10916-026-02469-z
Keywords: cardiopulmonary resuscitation, chest compressions, QCPR score, feedback system, compression depth, compression rate, chest stiffness, resuscitation training, manikin study, CPR guidelines, Laerdal, simulation
News Source: Ophelia Keating. (October 5, 2026). CPR Feedback Scores May Reward Compressions That Fall Outside Guidelines. Scienmag.
