Systematic development and validation of a predictive model for major postoperative complications in the Peri-operative Quality Improvement Project (PQIP) dataset

Complications are common following major surgery and are associated with increased use of healthcare resources, disability and mortality. Continued reliance on mortality estimates risks harming patients and health systems, but existing tools for predicting complications are unwieldy and inaccurate. We aimed to systematically construct an accurate pre-operative model for predicting major postoperative complications; compare its performance against existing tools; and identify sources of inaccuracy in predictive models more generally. Complete patient records from the UK Peri-operative Quality Improvement Programme dataset were analysed. Major complications were defined as Clavien-Dindo grade ≥ 2 for novel models. In a 75% train:25% test split cohort, we developed a pipeline of increasingly complex models, prioritising pre-operative predictors using Least Absolute Shrinkage and Selection Operators (LASSO). We defined the best model in the training cohort by the lowest Akaike's information criterion, balancing accuracy and simplicity. Of the 24,983 included cases, 6389 (25.6%) patients developed major complications. Potentially modifiable risk factors (pain, reduced mobility and smoking) were retained. The best-performing model was highly complex, specifying individual hospital complication rates and 11 patient covariates. This novel model showed substantially superior performance over generic and specific prediction models and scores. We have developed a novel complications model with good internal accuracy, re-prioritised predictor variables and identified hospital-level variation as an important, but overlooked, source of inaccuracy in existing tools. The complexity of the best-performing model does, however, highlight the need for a step-change in clinical risk prediction to automate the delivery of informative risk estimates in clinical systems.

Development and internal validation of a model for postoperative morbidity in adults undergoing major elective colorectal surgery: the peri-operative quality improvement programme (PQIP) colorectal risk model

Over 1.5 million major surgical procedures take place in the UK NHS each year and approximately 25% of patients develop at least one complication. The most widely used risk-adjustment model for postoperative morbidity in the UK is the physiological and operative severity score for the enumeration of mortality and morbidity. However, this model was derived more than 30 years ago and now overestimates the risk of morbidity. In addition, contemporary definitions of some model predictors are markedly different compared with when the tool was developed. A second model used in clinical practice is the American College of Surgeons National Surgical Quality Improvement Programme risk model; this provides a risk estimate for a range of postoperative complications. This model, widely used in North America, is not open source and therefore cannot be applied to patient populations in other settings. Data from a prospective multicentre clinical dataset of 118 NHS hospitals (the peri-operative quality improvement programme) were used to develop a bespoke risk-adjustment model for postoperative morbidity. Patients aged ≥ 18 years who underwent colorectal surgery were eligible for inclusion. Postoperative morbidity was defined using the postoperative morbidity survey at postoperative day 7. Thirty-one candidate variables were considered for inclusion in the model. Death or morbidity occurred by postoperative day 7 in 3098 out of 11,646 patients (26.6%). Twelve variables were incorporated into the final model, including (among others): Rockwood clinical frailty scale; body mass index; and index of multiple deprivation quintile. The C-statistic was 0.672 (95%CI 0.660-0.684), with a bootstrap optimism corrected C-statistic of 0.666 at internal validation. The model demonstrated good calibration across the range of morbidity estimates with a mean slope gradient of predicted risk of 0.959 (95%CI 0.894-1.024) with an index-corrected intercept of -0.038 (95%CI -0.112-0.036) at internal validation. Our model provides parsimonious case-mix adjustment to quantify risk of morbidity on postoperative day 7 for a UK population of patients undergoing major colorectal surgery. Despite the C-statistic of < 0.7, our model outperformed existing risk-models in widespread use. We therefore recommend application in case-mix adjustment, where incorporation into a continuous monitoring tool such as the variable life adjusted display or exponentially-weighted moving average-chart could support high-level monitoring and quality improvement of risk-adjusted outcome at the population level.

A learning health systems approach to improving the quality of care for patients in South Asia

Poor quality of care is a leading cause of excess morbidity and mortality in low- and middle- income countries (LMICs). Improving the quality of healthcare is complex, and requires an interdisciplinary team equipped with the skills to design, implement and analyse setting-relevant improvement interventions. Such capacity is limited in many LMICs. However, training for healthcare workers in quality improvement (QI) methodology without buy-in from multidisciplinary stakeholders and without identifying setting-specific priorities is unlikely to be successful. The Care Quality Improvement Network (CQIN) was established between Network for Improving Critical care Systems and Training (NICST) and University College London Centre for Perioperative Medicine, with the aim of building capacity for research and QI. A two-day international workshop, in collaboration with the College of Surgeons of Sri Lanka, was conducted to address the above deficits. Innovatively, the CQIN adopts a learning health systems (LHS) approach to improving care by leveraging information captured through the NICST electronic multi-centre acute and critical care surveillance platform. Fifty-two delegates from across the CQIN representing clinical, civic and academic healthcare stakeholders from six countries attended the workshop. Mapping of care processes enabled identification of barriers and drivers to the delivery of care and facilitated the selection of feasible QI methods and matrices. Six projects, reflecting key priorities for improving the delivery of acute care in Asia, were collaboratively developed: improving assessment of postoperative pain; optimising sedation in critical care; refining referral of deteriorating patients; reducing surgical site infection after caesarean section; reducing surgical site infection after elective general surgery; and improving provision of timely electrocardiogram recording for patients presenting with signs of acute myocardial infarction. Future project implementation and evaluation will be supported with resources and expertise from the CQIN partners. This LHS approach to building capacity for QI may be of interest to others seeing to improve care in LMICs.

Brinley plots, explained variance, and the analysis of age differences in response latencies

Critics of Brinley plot analyses (e.g., Fisk, Fisher, & Rogers, 1992; Perfect, 1994) claim: (a) that lack of overlap between latencies on different tasks inflates r2 values; (b) that Brinley plots mask task-specific age differences; and (c) that measurement error in young adults' latencies precludes the use of regression techniques with Brinley plot data. We dispute these claims. We show that lack of overlap does not inflate r2 and that the possible presence of task-specific effects in Brinley plot data may be evaluated using standard regression techniques. These techniques are illustrated using data from the Fisk and Rogers (1991) study of visual and memory search. Analysis of their data reveals significant differences between the lexical and nonlexical domains, but not between types of search. Finally, the effect of measurement error on Brinley plot analyses is shown to be small and, if taken into account, leads to increased support for general cognitive slowing.

Octreotide inhibits the meal-induced increases in the portal venous pressure of cirrhotic patients with portal hypertension: a double-blind, placebo-controlled study

The aim of this study was to determine the effects of the long-acting somatostatin analog, octreotide, on portal venous pressure and collateral blood flow in cirrhotic patients with portal hypertension during fasting and postprandial states. In a double-blind, placebo-controlled study, we investigated the effects of octreotide on the hepatic venous pressures and azygos blood flow of 21 patients before and after a standard liquid meal containing 40 gm of protein in 250 ml. Octreotide significantly reduced azygos blood flow from a mean of 499 +/- 65 ml/min to a mean of 355 +/- 47 ml/min (p < 0.01), but it had no effect on the hepatic venous pressure gradient. The hepatic venous pressure gradient of patients in the placebo group increased significantly, from a fasting mean of 16.4 +/- 1.6 mm Hg to a mean of 20.0 +/- 1.7 mm Hg 30 min after the meal (p < 0.01). In a second protocol hepatic venous pressures were measured in 20 patients at 30-min intervals for 2 hr after ingestion of the mixed meal. Again the placebo group showed a significant increase in the hepatic venous pressure gradient 30 min after the meal (20.4 +/- 1.5 mm Hg vs. 18.2 +/- 1.2 mm Hg; p < 0.05), but the group receiving octreotide showed no significant changes during the 2 hr of observation. We conclude that octreotide significantly reduces azygos blood flow, with little effect on portal venous pressure, and that it appears to inhibit postprandial increases in portal pressure in cirrhotic patients with portal hypertension.

The information-loss model: a mathematical theory of age-related cognitive slowing

A model of cognitive slowing is proposed with the following assumption: Information is lost during processing, processing occurs in discrete steps with step duration inversely related to the amount of information currently available, and the effect of aging is to increase the proportion of information lost per step. This model correctly predicts a positively accelerated relation between latencies of older and younger adults and provides a unified account of the effects of task complexity, practice, speed-accuracy tradeoffs, and fluctuations in individual performance. Strong support for the thesis that cognitive slowing is global, and not localized in specific age-sensitive components, is provided by the fact that the model accurately predicts the latencies of older adults on the basis of those of younger adults, without regard to the nature of the task, across a latency range of nearly 2 orders of magnitude.

The effect of non-protein liquid meals on the hepatic venous pressure gradient in patients with cirrhosis

It has been suggested that protein feeding increases portal pressure in cirrhotic patients, but that carbohydrate and fat have little effect. We examined the relationship between feeding and portal pressure, using different liquid test meals (250 or 500 ml non-protein, 250 ml protein-containing, 500 ml water), in 29 alcoholic patients with cirrhosis and portal hypertension. The mean hepatic venous pressure gradient (HVPG) increased significantly 30 min after the protein meal (10% increase; p = 0.009) and returned to basal levels at 60 min. The mean HVPG also increased significantly after the non-protein meal: after 500 ml the increase was 23% at 30 min (p = 0.046) and 17% at 60 min (p = 0.12); and after 250 ml it was 15% at 30 min (p = 0.012) and 7% at 60 min (p = 0.05). Ingestion of 500 ml water caused a small, non-significant, increase in mean HVPG. Plasma glucagon levels increased significantly at 30 and 60 min after the protein meal, but did not change significantly after the non-protein meal or water. Both protein-containing and non-protein meals significantly elevate HVPG in alcoholic patients with cirrhosis and portal hypertension.

General slowing of nonverbal information processing: evidence for a power law

Data were analyzed from studies of nonverbal information processing in which the dependent measure was the latency of pressing or releasing a response key. Positively-accelerated power functions described the relationship between the response latencies of groups of older (50 to 60 years and 65 to 75 years) and younger adults (20 to 25 years) with extreme precision (r2 = .99). The exponent of the best-fitting power function increased with the age of the older group. The form of the relationship is allometric, and is consistent with a model (Botwinick, 1984) in which response latency increases exponentially with task difficulty. The present findings suggest that this model holds across a wide variety of information-processing tasks and over a very broad range of latencies.

Long-term beneficial effects of endralazine, a new arteriolar vasodilator at rest and during exercise capacity in chronic congestive heart failure

Hemodynamic measurements were made at rest and during submaximal and maximal exercise in 12 patients with chronic congestive heart failure before and after oral endralazine (EN). After acute assessment, patients received endralazine, twice daily, for a mean of 2.8 months, when hemodynamic measurements were repeated. The drug was withdrawn for 3 to 4 days and subsequently reintroduced. Three patients with greatly elevated pulmonary wedge pressures were assessed after 30 mg of isosorbide dinitrate, which was also chronically administered. Resting mean cardiac and stroke volume indexes increased by 44 and 33%, respectively (p less than 0.01), with concomitant reduction of the systemic resistance. This improvement was maintained on a long-term basis in 8 of the 11 surviving patients. Withdrawal and subsequent reintroduction of EN confirmed that there was worsening of left ventricular dysfunction in the other 3 subjects. Chronic but not acute therapy produced a modest reduction in wedge pressure. At maximal exercise, cardiac and stroke volume indexes increased by 29 and 18%, respectively (p less than 0.01), after EN; the duration of exercise increased in 7 of the 10 subjects after acute therapy and this was maintained on a long-term basis. Mean creatinine clearance increased by 34% (p less than 0.01). The results confirm that EN produces acute and long-term hemodynamic and functional improvement without tolerance in congestive heart failure.

Acute hemodynamic effects of endralazine: a new vasodilator for chronic refractory congestive heart failure

Invasive hemodynamic measurements were made in 10 supine patients with chronic refractory congestive heart failure (CHF) due to ischemic heart disease or cardiomyopathy before and after oral administration of a new arteriolar vasodilator, endralazine. In 9 patients, a 10 mg dose of endralazine produced maximal increases in cardiac and stroke volume indexes of 56 and 41%, respectively, with a 45% reduction in total systemic resistance. After a 5 mg dose of endralazine, cardiac index increased maximally by 38% and stroke volume index by 34%, with a 31% decrease in total systemic resistance. Mean arterial pressure decreased 11 +/- 4 mm Hg (mean +/- standard error of mean) with the 5 mg dose and 17 +/- 5 mm Hg after the 10 mg dose. There were no significant changes in the right atrial, pulmonary arterial, or pulmonary capillary wedge pressures. After administration of a single dose of endralazine, statistically significant hemodynamic changes were observed from 1 to 8 hours with peak responses at 3 to 4 hours. These observations suggest that endralazine has hemodynamic properties similar to those of its structural analog, hydralazine. However, endralazine metabolism is largely independent of the patients' acetylator status, and no cases of systemic lupus erythematosus have been reported after long-term oral administration. These findings suggest that endralazine may be an efficacious drug that is potentially safer than hydralazine in the treatment of chronic CHF.

Comparison of serial electrocardiographic and vectorcardiographic changes during recovery from status asthmaticus

Serial electrocardiographs (ECG) and vectorcardiographs (VCG) have been performed on 10 patients admitted to hospital in status asthmaticus on 12 separate occasions. The VCG was more efficient than the ECG in the detection of right atrial and ventricular enlargement. Both investigations were equally reliable in recording changes in frontal plane P wave axis. The mean frontal plane P wave axis fell from +60 degrees (range +35 degrees to +90 degrees) on the day of admission to +43 degrees (range +30 degrees to +60 degrees) at the time of discharge. The mean FEV1 expressed as a percentage of predicted values increased from 48% (range 25% to 81%) to 87% (range 44% to 123%). In direct contrast to previous studies the presence of an abnormally vertical frontal plane P wave axis (> 60 degrees) was related to the severity of airway obstruction (p < 0.01).

Variations in Mechanical Sampling: Their Causes and Effects

Three characteristics of a sample of AutoAnalyzer Sampler II modules have been measured : eccentricity of the cam, eccentricity of the spindle driving the cam, and the lengths of the chords of the cam lobes. Imprecision in timing may make a significant contribution to analytical error.