Quality of Life After Cardiac Surgery: The QUACS Study and Benefit Prediction Model Authors: Samer A M Nashef, James Willard, Christine Mills, Yi-Da Chiu, Philip Noyes, Michael C Petch, Dominique-Laurent Couturier, Stephen R Large Published: 2026 ABSTRACT AbstractObjectivesSafer cardiac surgery is being offered to more high-risk and frail patients, improving both survival and quality of life (QoL) for most. The QoL After Cardiac Surgery Study (QUACS) evaluated QoL changes postoperatively to identify those who do not benefit from and may be harmed by intervention.MethodsModerate- and high-risk cardiac surgery patients enrolled in a prospective, national multicentre study (2019-2023). Mortality risk, frailty, outlook, affect, and socio-economic status were objectively measured. Quality of life was assessed preoperatively and to 12-24 months postoperatively.ResultsParticipants were 2948 patients (EuroSCORE II ≥3 or logistic EuroSCORE ≥6, mean age 72.9 years, female 39.6%). Inclusion criteria were not met in 186 patients, leaving 2762 for analysis. Hospital mortality was 3.93% and 5.86% died after discharge. At 1 year after surgery, net QoL benefit was achieved in 74% of symptomatic patients and in 18% of patients thought asymptomatic. Risk factors for failure to achieve a net QoL benefit were identified, and a preoperative QoL calculator was developed to quantify the likelihood of QoL benefit.ConclusionsThis first study examining QoL changes over time after complex major intervention such as cardiac surgery confirms that some patients do not obtain QoL benefit and helps to identify them. Findings and methodology are applicable to a wide range of major operations. The QUACS Study will improve the process of informed consent and decision-making by patients contemplating such procedures and reduce harmful interventions.Clinical Trial Registration NumberNCT04231461 (https://clinicaltrials.gov/study/NCT04231461). Introduction During the last 20 years, heart surgery has become much safer, and more frail, high-risk, and elderly patients are being offered increasingly complex open-heart surgery for various heart conditions. Heart operations usually improve both long-term survival and quality of life (QoL), but that is not true for all. In a small percentage of patients, a heart operation can abruptly terminate life and, in some patients, QoL may be adversely affected in the long term. In terms of survival, risk models such as EuroSCORE II1 can calculate the risk to life from having a heart operation quite precisely. Knowledge of the natural history of the cardiac condition can determine the risk to life over time from not having surgery for most heart conditions and most patients. The immediate impact of a heart operation on survival is negative. Depending on the clinical situation, once patients survive surgery, most stand to benefit in long-term survival. The survival benefit is well documented for most heart conditions, and the length of time that must pass before patients begin to achieve a net survival benefit can be clearly defined and measured in advance, and this has been addressed in our previous work.2 QoL follows a similar pattern to survival: it worsens immediately after surgery then, in most patients, recovers to baseline and continues to improve so that, once QoL loss from surgery is compensated, the patient begins to achieve a net QoL benefit from that time onwards. However, we have very little knowledge about the net impact of heart operations on QoL, which is the outcome that many patients care about most of all. We also know that 8%-19% of octogenarian patients derive no QoL benefit3 after cardiac surgery but have no way of identifying them before harm is done. Although some studies of cardiac surgery have included QoL as an outcome measure, no study has ever analysed the changes in QoL over time in the postoperative period, and no study has ever analysed the relationship between individual patient and operation characteristics and the QoL outcome. To provide true and fully informed consent, patients need to know the impact of surgery on both survival and QoL after surgery. This study was designed to provide this information, so that patients are empowered to give evidence-based and truly informed consent when they are contemplating a major heart operation. This study was designed to address the following important questions: How does having heart surgery affect QoL after the operation? How does having heart surgery affect QoL after the operation? How long does it take patients to return to the same QoL they had before surgery? How long does it take patients to return to the same QoL they had before surgery? Does QoL improve thereafter? If so, to what extent? Does QoL improve thereafter? If so, to what extent? How long does it take patients to regain any QoL lost due to the operation? How long does it take patients to regain any QoL lost due to the operation? What proportion of patients will achieve a net benefit in QoL? If so, when? What proportion of patients will achieve a net benefit in QoL? If so, when? Are there factors that can help predict who will benefit in terms of QoL and who will not? Are there factors that can help predict who will benefit in terms of QoL and who will not? Methods From July 2019 to February 2023, in a nationwide, UK multicentre (29 hospitals) prospective observational study, patients having elective or urgent major heart surgery were offered participation if they were deemed to be at moderate or high risk, defined as having a EuroSCORE II ≥ 31 or a logistic EuroSCORE ≥ 64 (Figure 1). A sample size calculation was performed, and enough patients were recruited to ensure adequate predictive power of the developed QoL calculator (QoLculator; details are provided in Sections S5.1 and S5.2 of Appendix B). All major cardiac operations were included, such as coronary bypass, heart valve procedures, surgery on the thoracic aorta, and any combination of the above. Lay patient representation was consulted before the final study design. All participants gave fully informed written consent. CONSORT Diagram of the QUACS Study. Levels of completeness of QoL data at each time point during the first and second years of post-surgery follow-up. Reasons for missingness include death, withdrawal of consent (abbreviated “WoC”), loss to follow-up (abbreviated “LFu”), and noncompletion of a questionnaire at a given time point while still alive and contributing to the study (column “Missing”). (*) For death, withdrawal of consent and loss to follow-up, cumulative numbers over the period of interest are provided. Abbreviations: LFu, loss to follow-up; QoL, quality of life; WoC, withdrawal of consent. Data were collected on demographics and risk profile (Table 1), operation types, complexity, and bypass and aortic cross-clamp times (Table 2), and pre-operative QoL status (Table 3). Intensive care and hospital stay, hospital mortality, and survival after discharge were also recorded. Table 1.Patient Demographics and Risk ProfileSymptomatic EQ5-TAsymptomatic EQ5-TOverall(N = 2342)(N = 420)(N = 2762)Age (years) Mean (SD)72.8 (9.39)73.7 (9.92)72.9 (9.48) Median [IQR]75.2 [9.75]75.9 [9.17]75.3 [9.55]Sex Female937 (40.0%)156 (37.1%)1093 (39.6%) Male1405 (60.0%)264 (62.9%)1669 (60.4%)Height (m) Mean (SD)1.69 (0.103)1.69 (0.109)1.69 (0.104) Median [IQR]1.70 [0.140]1.69 [0.160]1.70 [0.160] Missing11 (0.5%)1 (0.2%)12 (0.4%)Weight (kg) Mean (SD)79.9 (17.1)76.9 (15.3)79.5 (16.9) Median [IQR]78.4 [22.0]76.4 [21.0]78.0 [23.0] Missing2 (0.1%)0 (0%)2 (0.1%)BMI (kg/m²) Mean (SD)28.0 (5.29)26.9 (4.61)27.8 (5.20) Median [IQR]27.3 [7.00]26.5 [5.60]27.2 [6.70] Missing11 (0.5%)1 (0.2%)12 (0.4%)Education ≤GCSE981 (41.9%)173 (41.2%)1154 (41.8%)≥A-level843 (36.0%)157 (37.4%)1000 (36.2%) ≥BA476 (20.3%)89 (21.2%)565 (20.5%) Missing42 (1.8%)1 (0.2%)43 (1.6%)Ethnicity White2230 (95.2%)404 (96.2%)2634 (95.4%) Non-White112 (4.8%)16 (3.8%)128 (4.6%)Employment Inactive1973 (84.2%)350 (83.3%)2323 (84.1%) Active365 (15.6%)70 (16.7%)435 (15.7%) Missing4 (0.2%)0 (0%)4 (0.1%)Total frailty score 0527 (22.5%)122 (29.0%)649 (23.5%) 1534 (22.8%)112 (26.7%)646 (23.4%) 2705 (30.1%)120 (28.6%)825 (29.9%) 3344 (14.7%)35 (8.3%)379 (13.7%) 4114 (4.9%)10 (2.4%)124 (4.5%) 51 (0.0%)0 (0%)1 (0.0%) Missing117 (5.0%)21 (5.0%)138 (5.0%)EuroSCORE II Mean (SD)5.43 (3.41)5.31 (2.98)5.41 (3.35) Median [IQR]4.32 [2.43]4.23 [2.48]4.31 [2.43] Missing1 (0.0%)0 (0%)1 (0.0%)Extracardiac arteriopathy No2027 (86.6%)380 (90.5%)2407 (87.1%) Yes315 (13.5%)40 (9.5%)355 (12.9%)Neurological dysfunction No2298 (98.1%)416 (99.0%)2714 (98.3%) Yes44 (1.9%)4 (1.0%)48 (1.7%)Poor mobility No2226 (95.0%)414 (98.6%)2640 (95.6%) Yes116 (5.0%)6 (1.4%)122 (4.4%)Previous cardiac surgery No2002 (85.5%)359 (85.5%)2361 (85.5%) Yes339 (14.5%)61 (14.5%)400 (14.5%) Missing1 (0.0%)0 (0%)1 (0.0%)Chronic lung disease No1930 (82.4%)372 (88.6%)2302 (83.3%) Yes412 (17.6%)48 (11.4%)460 (16.7%)Active endocarditis No2273 (97.1%)409 (97.4%)2682 (97.1%) Yes69 (2.9%)11 (2.6%)80 (2.9%)Critical preoperative state No2317 (98.9%)419 (99.8%)2736 (99.1%) Yes25 (1.1%)1 (0.2%)26 (0.9%)Diabetes on insulin No2194 (93.7%)411 (97.9%)2605 (94.3%) Yes148 (6.3%)9 (2.1%)157 (5.7%)Dialysis No2317 (98.9%)418 (99.5%)2735 (99.0%) Yes25 (1.1%)2 (0.5%)27 (1.0%)NYHA I460 (19.6%)138 (32.9%)598 (21.7%) II960 (41.0%)216 (51.4%)1176 (42.6%) III768 (32.8%)56 (13.3%)824 (29.8%) IV99 (4.2%)3 (0.7%)102 (3.7%) Missing55 (2.3%)7 (1.7%)62 (2.2%)CCS 01424 (60.8%)303 (72.1%)1727 (62.5%) 1284 (12.1%)48 (11.4%)332 (12.0%) 2330 (14.1%)41 (9.8%)371 (13.4%) 3173 (7.4%)13 (3.1%)186 (6.7%) 488 (3.8%)11 (2.6%)99 (3.6%) Missing43 (1.8%)4 (1.0%)47 (1.7%)Unstable angina No2282 (97.4%)412 (98.1%)2694 (97.5%) Yes55 (2.3%)7 (1.7%)62 (2.2%) Missing5 (0.2%)1 (0.2%)6 (0.2%)LV function Good (LVEF >50%)1434 (61.2%)270 (64.3%)1704 (61.7%) Moderate (LVEF 31-50%)661 (28.2%)121 (28.8%)782 (28.3%) Poor (LVEF 21-30%)180 (7.7%)20 (4.8%)200 (7.2%) Very poor (LVEF ≤ 20%)40 (1.7%)4 (1.0%)44 (1.6%) Missing27 (1.2%)5 (1.2%)32 (1.2%)Recent MI No1940 (82.8%)350 (83.3%)2290 (82.9%) Yes400 (17.1%)70 (16.7%)470 (17.0%) Missing2 (0.1%)0 (0%)2 (0.1%)Urgency Elective1486 (63.5%)284 (67.6%)1770 (64.1%) Urgent855 (36.5%)136 (32.4%)991 (35.9%) Missing1 (0.0%)0 (0%)1 (0.0%)Plasma creatinine (µmol/L) Mean (SD)97.9 (60.6)92.8 (56.7)97.2 (60.0) Median [IQR]87.0 [33.0]87.0 [29.0]87.0 [33.0] Missing2 (0.1%)2 (0.5%)4 (0.1%)Serum albumin (g/L) Mean (SD)39.8 (5.32)39.8 (5.04)39.8 (5.28) Median [IQR]40.0 [6.00]40.5 [6.00]40.0 [6.00] Missing91 (3.9%)18 (4.3%)109 (3.9%)Haemoglobin (g/L) Mean (SD)130 (17.9)133 (16.6)131 (17.7) Median [IQR]132 [23.3]133 [21.0]132 [23.0] Missing2 (0.1%)1 (0.2%)3 (0.1%)Abbreviations: MI: myocardial infarction. LV: left ventricle. LVEF: LV ejection fraction. EQ5-T: EuroQol5 - health thermometer. EQ5-U: EuroQol - utility. Patient Demographics and Risk Profile Abbreviations: MI: myocardial infarction. LV: left ventricle. LVEF: LV ejection fraction. EQ5-T: EuroQol5 - health thermometer. EQ5-U: EuroQol - utility. Table 2.Surgical Procedures, Intraoperative Times, and OutcomesSymptomatic EQ5-TAsymptomatic EQ5-TOverall(N = 2342)(N = 420)(N = 2762)Weight of the intervention Isolated CABG399 (17.0%)61 (14.5%)460 (16.7%) Single non-CABG594 (25.4%)121 (28.8%)715 (25.9%) 2 procedures992 (42.4%)180 (42.9%)1172 (42.4%) 3 or more procedures356 (15.2%)58 (13.8%)414 (15.0%) Missing1 (0.0%)0 (0%)1 (0.0%)Surgery on the thoracic aorta No1889 (80.7%)309 (73.6%)2198 (79.6%) Yes446 (19.0%)111 (26.4%)557 (20.2%) Missing7 (0.3%)0 (0%)7 (0.3%)Died in hospital No2226 (95.0%)393 (93.6%)2619 (94.8%) Yes91 (3.9%)16 (3.8%)107 (3.9%) Missing25 (1.1%)11 (2.6%)36 (1.3%)Consented to extended follow-up No305 (13.0%)67 (16.0%)372 (13.5%) Yes1272 (54.3%)229 (54.5%)1501 (54.3%) Missing765 (32.7%)124 (29.5%)889 (32.2%)ProcedureCABG No1261 (53.8%)235 (56.0%)1496 (54.2%) Yes1081 (46.2%)185 (44.0%)1266 (45.8%)Aortic valve No1095 (46.8%)191 (45.5%)1286 (46.6%) Yes1247 (53.2%)229 (54.5%)1476 (53.4%)Mitral valve No1752 (74.8%)323 (76.9%)2075 (75.1%) Yes590 (25.2%)97 (23.1%)687 (24.9%)Tricuspid valve No2104 (89.8%)387 (92.1%)2491 (90.2%) Yes238 (10.2%)33 (7.9%)271 (9.8%)Aorta No1899 (81.1%)337 (80.2%)2236 (81.0%) Yes443 (18.9%)83 (19.8%)526 (19.0%)Other No1948 (83.2%)350 (83.3%)2298 (83.2%) Yes394 (16.8%)70 (16.7%)464 (16.8%)Bypass No67 (2.9%)20 (4.8%)87 (3.1%) Yes2273 (97.1%)400 (95.2%)2673 (96.8%) Missing2 (0.1%)0 (0%)2 (0.1%)Bypass time (min) Mean (SD)139 (68.6)139 (70.1)139 (68.8) Median [IQR]123 [78.0]125 [72.0]123 [78.0] Missing82 (3.5%)22 (5.2%)104 (3.8%)Cross clamp No56 (2.4%)6 (1.4%)62 (2.2%) Yes2213 (94.5%)395 (94.0%)2608 (94.4%) Missing73 (3.1%)19 (4.5%)92 (3.3%)Cross clamp time (min) Mean (SD)99.1 (51.1)100 (52.3)99.2 (51.3) Median [IQR]88.0 [59.0]89.0 [65.0]88.0 [59.0] Missing146 (6.2%)27 (6.4%)173 (6.3%)Abbreviation: EQ-5T: EuroQol 5 - health thermometer. CABG: coronary artery bypass grafting. SD: standard deviation. IQR: interquartile range. Surgical Procedures, Intraoperative Times, and Outcomes Abbreviation: EQ-5T: EuroQol 5 - health thermometer. CABG: coronary artery bypass grafting. SD: standard deviation. IQR: interquartile range. Table 3.Patient Baseline Quality of Life CharacteristicsSymptomatic EQ5-TAsymptomatic EQ5-TOverall(N = 2342)(N = 420)(N = 2762)EQ5-T Mean (SD)56.0 (19.0)91.0 (4.65)61.3 (21.6) Median [IQR]60.0 [25.0]90.0 [5.00]65.0 [30.0]EQ5-U Mean (SD)1.27 (0.245)1.46 (0.138)1.30 (0.242) Median [IQR]1.30 [0.243]1.48 [0.195]1.34 [0.271] Missing16 (0.7%)1 (0.2%)17 (0.6%)LOT Mean (SD)16.0 (5.14)17.5 (4.74)16.3 (5.11) Median [IQR]16.0 [8.00]18.0 [8.00]16.0 [7.00] Missing2 (0.1%)3 (0.7%)5 (0.2%)PANAS + Mean (SD)30.2 (8.31)34.2 (8.23)30.8 (8.43) Median [IQR]30.0 [12.0]35.0 [11.0]31.0 [12.0] Missing21 (0.9%)2 (0.5%)23 (0.8%)PANAS− Mean (SD)20.1 (7.45)16.6 (5.98)19.6 (7.35) Median [IQR]19.0 [11.0]15.0 [7.00]18.0 [10.0] Missing21 (0.9%)2 (0.5%)23 (0.8%)Abbreviations: EQ5-T EuroQoL5 - health thermometer. EQ5-U: EuroQol5 - utility. SD: standard deviation. IQR: interquartile range. Patient Baseline Quality of Life Characteristics Abbreviations: EQ5-T EuroQoL5 - health thermometer. EQ5-U: EuroQol5 - utility. SD: standard deviation. IQR: interquartile range. At baseline assessment, QoL was measured using EuroQoL-5D5 using EQ5 health thermometer (EQ5-T) and EQ5 utility (EQ5-U). In addition, affect was measured using positive and negative affect scales6 (PANAS) and outlook was measured using the life orientation test7 (LOT). Frailty was evaluated by the sit-to-stand test.8 After surgery, QoL was measured on discharge from hospital and monthly thereafter for 12 months (online and telephone). Patients were then offered the option of extended follow-up for a further 12 months of quarterly QoL assessment. In terms of QoL, “treatment failure” was pre-defined as death in hospital, death after discharge and before return to baseline QoL, failure to return to baseline QoL, or not achieving sufficient improvement in QoL to compensate for the loss incurred by having the operation. “Treatment failure” here refers purely to not achieving a net QoL benefit after surgery and is not a reflection on the technical success of an operation, or survival, or achieving prognostic benefit. Patients with a normal preoperative QoL (≥ 85) EQ5-T score before surgery were deemed to be having surgery on prognostic grounds alone and were never expected to obtain QoL benefit from surgery. They are referred to throughout as “asymptomatic,” and results are presented separately for the symptomatic and asymptomatic patient subgroups. Statistical analysis Because immediate post-surgery QoL cannot be measured, 3 assumptions were made about immediate postoperative drops in QoL from baseline: an absolute drop equal to 25% of the maximum score, and relative drops of 25% and 50% from each patient’s baseline score (Figure 2). There was very little difference on result analysis between the 3 assumptions, and a relative drop of 50% was adopted as the most conservative approach. Typical Patient’s QoL Trajectory After Cardiac Surgery. The solid grey line shows EQ5-T QoL status over time assuming 3 different reductions in QoL immediately after surgery (coloured dashed lines). The red shaded area represents surgery-related QoL loss or the “cost” of surgery in terms of QoL. T1 is the time at which the patient returns to baseline QoL. The green shaded area represents improvement over baseline QoL. T2 is the time at which the green shaded area equals the red shaded area and corresponds to the moment the surgery-related QoL loss has been compensated for by QoL improvement. Abbreviations: EQ5-T, EQ5 health thermometer; QoL, quality of life. For every patient, the surgery-related QoL loss was the area under the QoL curve (AUC) from surgery until the first time (T1) when the patient returned to baseline QoL (red area in Figure 2). For each QoL measure type (EQ5-T and EQ5-U), time to net QoL benefit (T2) was defined as the time when the loss in QoL induced by surgery was fully compensated. T2 occurred once a positive AUC (green area in Figure 2) equalled the surgery-related AUC initial QoL loss (red area in Figure 2). Areas under the QoL curve were estimated in 2 ways: parametrically, using random change-point mixed models, and non-parametrically using the trapezoidal rule. The times to net QoL benefit obtained by both methods were strongly correlated (Figure S1). The time needed to reach baseline QoL (T1) was determined by linear interpolation. The analysis variant considering the parametric modelling of longitudinal QoL measures allowed T1 and T2 to be estimated in the presence of right censoring (caused by loss of follow-up, withdrawal of consent, or a last QoL measure occurring before study end). The binary end-point, treatment QoL benefit, was defined as “success” if a net QoL benefit occurred before death or study end (whichever came first), and “failure” otherwise. Missingness in the treatment benefit end-point and predictors was handled via multiple imputation by chained equations which used predictive mean matching to produce 100 imputed datasets. Analyses were performed on each dataset and then pooled to obtain estimates, such as the percentage of patients with treatment benefit by QoL measure (EQ5-T and EQ5-H), at specific post-surgery time windows (12, 18, and 24 months), and subgroup (symptomatic vs asymptomatic). A predictive model for treatment benefit was developed using various machine learning classifiers, including random forest, boosting, and variants of logistic regression. The algorithms considered 35 predictors, including demographic, clinical, and baseline QoL measures (full list in Section S7.3 of Appendix A). Eighty percent of patients were randomly selected to define the training set common to all imputed datasets. The remaining 20% of patients formed the test set. On each imputed training set, models were optimized by K-fold cross-validation. Classification performance of the selected classifier and model was then assessed using the area under the receiver operating characteristic curve (AUROC) estimated on each imputed test set. All analyses were performed in R version 4.5.2. Results Participants were 2948 patients with EuroSCORE II ≥3 or logistic EuroSCORE ≥6 from 29 cardiac surgery centres. In-hospital mortality was 3.93%. Overall mortality rose to 7.78% at 360 days, 8.79% at 540 days, and 9.79% at 720 days. Minimum inclusion criteria (including absence of baseline QoL measures or surgical procedure) were not met in 186 patients (excluded), leaving 2762 patients for analysis. This exceeds the sample size of 2599 identified as the number of patients needed to develop a prediction model for treatment benefit, as per our planned sample size re-estimation (Section S5.2 of Appendix A). Mean age was 72.9 years (range 18-89, SD 9.5) and 39.6% were female. Mean EuroSCORE II was 5.4 (SD 3.4). Demographics, risk profile, and surgical data are in Table 1. Mean QoL improves after the first (lowest) postoperative measure. On average, patients reach T1, their preoperative baseline QoL, between 55 days (EQ5-T) or 80 days (EQ5-U) when assuming a relative QoL decrease of 50% at time of surgery. Average QoL continues to rise afterwards, so that surgery-related QoL loss is reversed by 93 days (EQ5-T) to 121 days (EQ5-U) on average, at which point (T2) net QoL benefit can begin (Figures 2 and 3). Mean Change in QoL from the First Postoperative Measure for All Patients. Quality of life (y-axis) as a function of time (x-axis) after surgery (time 0) for all patients (individual grey lines) according to 2 QoL types: EQ-5D health thermometer (EQ5-T, upper plot) and utility (EQ5-U, lower plot). The brown line shows the average trend over all patients. On average for our cohort, QoL increases sharply from the first post-operative measurement during the first quarter, then improvement slows and a plateau is reached. As a composite end-point defined on 5 questions, EQ5-U shows less variability than EQ5-T, which is defined by a single patient input. Abbreviations: EQ5-T, EQ5 health thermometer; EQ5-U, EQ5 utility. Improvement continues and reaches a plateau at 6 months, remaining largely static up to 2 years (Figure 3). EQ5 health thermometer QoL treatment benefit is reached by 65%, 73%, and 77% of all patients at 1 year, 18 months, and 2 years, respectively (Figure S2B), leaving 23% of all patients with no QoL improvement 2 years after surgery. Primary outcome By patient subgroup, net QoL benefit at 1 year was achieved in 74% of symptomatic patients and in 18% of asymptomatic patients. For asymptomatic patients, QoL was not the surgical indication, and surgery was largely for prognosis. Nevertheless, 18% had actually gained net QoL benefit at 1 year. For symptomatic patients, this leaves 26% in whom surgery did not achieve hoped-for net QoL benefit at 1 year. Secondary outcome In the 55% of patients who consented to extended follow-up, net QoL benefit at 18 months and 2 years was achieved in 34% and 43% of asymptomatic patients and in 80% and 82% of symptomatic patients. For patients deemed asymptomatic, QoL was not the surgical indication. Nevertheless, 43% had actually gained net QoL benefit by 2 years. For symptomatic patients, this leaves 18% of in whom surgery did not achieve hoped-for net QoL benefit at 2 years (Figure S3B). A comparison of the K-fold AUROC estimates of the different machine learning classifiers aiming to best predict treatment benefit in the training set shows that methods based on random forest, generalized linear model and vector-generalized linear models led to similar discriminative performance (Figures S4A and B). Logistic regression with stepwise model selection based on the Akaike information criterion was selected as the preferred method, as it was the simplest and most interpretable final model. The stepwise model selection process led to the identification of 17 predictors for treatment benefit. Unsurprisingly, preoperative symptom severity and low baseline QoL were strong predictors of postoperative QoL benefit (Figure 4). The final model was then validated on the test set, achieving an AUROC of 0.83 at 1 year after surgery (Figure 5). As such, it is used as the basis for the QoL calculator. A sensitivity analysis that defined treatment benefit using non-parametric estimates of time to net benefit (T2) and that did not perform multiple imputation led to an AUROC of 0.80 on the test set, suggesting that results are robust to the assumptions made by the random change point models and multiple imputation. Forest Plot of the Training Set Regression Coefficient Estimates (Points, Ordered from Smallest to Largest) and Related 95% Confidence Intervals (Bars) on the Log-odds Scale When Predicting Treatment Benefit as Defined by EQ5-T QoL Measures Using the Selected Multiple Logistic Regression Model. For continuous variables, estimates represent the change in the log-odds of treatment benefit associated with a 1 standard deviation unit increase from the mean. For categorical variables, the estimates represent the change in log-odds for treatment benefit when comparing the level listed on the left-hand side y-axis to its referent category. Negative (positive) values of the log-odds indicate a decreased (increased) probability of treatment benefit. Scaling and referent category information is provided in Appendix. Abbreviations: MI: myocardial infarction. LV: left ventricle. LVEF: LV ejection fraction. EQ5-T: EuroQol5 - health thermometer. EQ5-U: EuroQol - utility. Receiver Operating Characteristic (AUROC) Curves for the Training and Testing Datasets Using the Selected Multiple Logistic Regression Model. Abbreviation: AUROC, area under the receiver operating characteristic curve. Discussion Cardiac surgery has a positive impact on survival and quality of life in most patients. The survival impact is well documented, and prior studies have assessed QoL after cardiac surgery,9,10 but there has been no objective documentation of QoL changes over time. The only study to investigate risk factors with an impact on QoL after cardiac surgery is a recent, single-hospital study.11 Our study has documented the QoL changes after cardiac surgery over time, confirmed the expected immediate QoL loss and shown that in 77% of all patients that loss is “repaid” and that there is a net QoL benefit. Therefore, cardiac surgery produces substantial net QoL improvement in more than 3 quarters of moderate- to high-risk patients. This leaves 23% of patients without net benefit in QoL. There are 2 important findings in our QoL measurements. The first is that, in patients with a normal preoperative QoL who believed themselves asymptomatic, 18% nevertheless achieved a net QoL benefit at 1 year. This accords with the experience of cardiac surgeons, who often see postoperative patients declare that they feel better after surgery, despite claiming a normal QoL before surgery, presumably having been unaware of a slow QoL drop or ascribing it to the expected impact of ageing. The second is that in symptomatic patients with impaired preoperative QoL, surgery was partly or wholly indicated to alleviate symptoms and improve QoL and, at 1 year, 26% of these patients had not achieved the hoped-for benefit. Failure to achieve a net QoL benefit may be due either to unsuccessful treatment against a high-risk background or, in some patients, to a high or normal preoperative QoL that is difficult to improve on. Either way, such patients should only be offered surgery if there is a robust prognostic indication. Our QoLculator helps to identify these patients before surgery so that fully informed consent can be obtained. “Do no harm” is the fundamental guiding principle in all medical care. Every therapeutic intervention has benefits and risks. Properly informed consent cannot take place without communicating to the patient the likely magnitude of the benefits and risks of an intervention. As with all medical treatment, cardiac surgery aims to enhance survival, or QoL, or preferably both. With the advent of cardiac surgical risk models, it is now possible to quantify the risk to life of a heart operation and to compare it to the risk to life of not having a heart operation over time.2 The risk to life is immediately increased by surgery but, in most life-threatening cardiac surgical conditions, that risk falls with time after surgery and survival begins to exceed that of conservative therapy. The same is true of QoL but, until now, the change in QoL over time after a heart operation and the risk factors associated with lack of QoL benefit have neither been described nor quantified. The QoL After Cardiac Surgery Study (QUACS) Study corrects this deficiency and provides an evidence base and a predictive model for QoL which is similar to what EuroSCORE supplies for survival. Limitations It is possible that the Covid 19 pandemic may have somewhat distorted QoL reporting in some patients who may have contracted the virus during the time of QoL measurements. The impact of the pandemic on cardiac surgical throughput in the UK meant that the study took longer to complete than planned. Extended follow-up beyond 1 year was entirely voluntary. The number of patients who consented to extended follow-up was 1507, which is about 55% of the 2778 patients who consented and were eligible at the beginning of the study (Figure 1). It is also possible that consenting to extended follow-up may have been affected by the QoL status at the time, so the 2-year data should be interpreted with caution. Our assumption of a 50% drop in QoL immediately after surgery was empirical in that this assumption, while robust in all sensitivity analyses, was not directly measured and may vary by procedure or complication profile. The methodology of this study can be adopted for major interventions in other specialties, but this may be limited by healthcare system factors and rehabilitation pathways. Cultural perceptions of QoL may influence generalizability outside the UK Conclusion Despite claiming a normal preoperative QoL, 18% of asymptomatic patients achieve net QoL benefit at 1 year, whereas, in symptomatic patients, 74% will have achieved net QoL benefit at 1 year leaving 26% in whom net QoL benefit has still not been achieved. The QoLculator will allow evaluation of the likelihood of obtaining QoL benefit, and we recommend that it is used in preoperative assessment and as part of the process of seeking informed consent for surgery. Supplementary Material