Accurate Patient-Specific Machine Learning Models of Glioblastoma Invasion Using Transfer Learning.

BACKGROUND AND PURPOSE: MR imaging-based modeling of tumor cell density can substantially improve targeted treatment of glioblastoma. Unfortunately, interpatient variability limits the predictive ability of many modeling approaches. We present a transfer learning method that generates individualized patient models, grounded in the wealth of population data, while also detecting and adjusting for interpatient variabilities based on each patient's own histologic data. MATERIALS AND METHODS: We recruited patients with primary glioblastoma undergoing image-guided biopsies and preoperative imaging, including contrast-enhanced MR imaging, dynamic susceptibility contrast MR imaging, and diffusion tensor imaging. We calculated relative cerebral blood volume from DSC-MR imaging and mean diffusivity and fractional anisotropy from DTI. Following image coregistration, we assessed tumor cell density for each biopsy and identified corresponding localized MR imaging measurements. We then explored a range of univariate and multivariate predictive models of tumor cell density based on MR imaging measurements in a generalized one-model-fits-all approach. We then implemented both univariate and multivariate individualized transfer learning predictive models, which harness the available population-level data but allow individual variability in their predictions. Finally, we compared Pearson correlation coefficients and mean absolute error between the individualized transfer learning and generalized one-model-fits-all models. RESULTS: Tumor cell density significantly correlated with relative CBV (r = 0.33, P < .001), and T1-weighted postcontrast (r = 0.36, P < .001) on univariate analysis after correcting for multiple comparisons. With single-variable modeling (using relative CBV), transfer learning increased predictive performance (r = 0.53, mean absolute error = 15.19%) compared with one-model-fits-all (r = 0.27, mean absolute error = 17.79%). With multivariate modeling, transfer learning further improved performance (r = 0.88, mean absolute error = 5.66%) compared with one-model-fits-all (r = 0.39, mean absolute error = 16.55%). CONCLUSIONS: Transfer learning significantly improves predictive modeling performance for quantifying tumor cell density in glioblastoma.

Coronary and cardiovascular risk estimation for primary prevention: validation of a new Sheffield table in the 1995 Scottish health survey population.

OBJECTIVE: To examine the accuracy of a new version of the Sheffield table designed to aid decisions on lipids screening and detect thresholds for risk of coronary heart disease needed to implement current guidelines for primary prevention of cardiovascular disease. DESIGN: Comparison of decisions made on the basis of the table with absolute risk of coronary heart disease or cardiovascular disease calculated by the Framingham risk function. The decisions related to statin treatment when coronary risk is >/=30% over 10 years; aspirin treatment when the risk is >/=15% over 10 years; and the treatment of mild hypertension when the cardiovascular risk is >/=20% over 10 years. SETTING: The table is designed for use in general practice. SUBJECTS: Random sample of 1000 people aged 35-64 years from the 1995 Scottish health survey. MAIN OUTCOME MEASURES: Sensitivity, specificity, and positive and negative predictive values of the table. RESULTS: 13% of people had a coronary risk of >/=15%, and 2. 2% a risk of >/=30%, over 10 years. 22% had mild hypertension (systolic blood pressure 140-159 mm Hg). The table indicated lipids screening for everyone with a coronary risk of >/=15% over 10 years, for 95% of people with a ratio of total cholesterol to high density lipoprotein cholesterol of >/=8.0, but for <50% with a coronary risk of <5% over 10 years. Sensitivity and specificity were 97% and 95% respectively for a coronary risk of >/=15% over 10 years; 82% and 99% for a coronary risk of >/=30% over 10 years; and 88% and 90% for a cardiovascular risk of >/=20% over 10 years in mild hypertension. CONCLUSION: The table identifies all high risk people for lipids screening, reduces screening of low risk people by more than half, and ensures that treatments are prescribed appropriately to those at high risk, while avoiding inappropriate treatment of people at low risk.

The rationale for differing national recommendations for the treatment of hypertension.

This article examines the rationale for the differences in the guidelines for hypertension management of four national or international bodies: the Joint National Committee (JNC-V), The World Health Organization/International Society of Hypertension (WHO-ISH), the British Hypertension Society (BHS), and the New Zealand guidelines. These guidelines agree on many aspects of management, but differ on two very important points-the drugs of first choice for hypertension, and the indications for drug treatment of uncomplicated mild hypertension. JNC-V recommends treatment routinely of all people with a sustained blood pressure of 140/90 mm Hg, whereas the BHS guidelines advise treatment routinely at 160/100 mm Hg. Such differences in the threshold for treatment have a major impact on the proportion of the adult population to be treated, and on the benefit from treatment. JNC-V was heavily influenced by the Hypertension Detection and Follow-up Program (HDFP), which appeared to show a large benefit from the treatment of uncomplicated mild hypertension, whereas the BHS guidelines were influenced by the Medical Research Council (MRC) Trial, which showed a very small benefit. However, the apparent differences in absolute benefit between these, and other, randomized controlled trials is related entirely to differences in the absolute cardiovascular risk of the populations studied. In populations and in individual patients the benefit from antihypertensive treatment is determined by the absolute cardiovascular risk. Blood pressure by itself is a very weak predictor of risk or benefit from treatment. In uncomplicated mild hypertension the need for drug therapy should be based on the absolute risk of cardiovascular complications, estimated by considering age, sex, serum cholesterol level, diabetes mellitus status, and smoking habits, in addition to blood pressure. Doctors cannot estimate absolute risk accurately informally or intuitively, and the next generation of guidelines should incorporate a simple but accurate method for estimating cardiovascular risk, similar to that in the New Zealand guidelines. The decision to treat, or not treat, uncomplicated mild hypertension should be based on a formal estimate of absolute cardiovascular risk and not on an arbitrary blood pressure threshold. As regards drugs of first choice, the available evidence supports strongly the stance of JNC-V and JNC VI that diuretics and beta-blockers should be preferred unless they are contraindicated, or unless there are positive indications for other drug classes.

Sheffield risk and treatment table for cholesterol lowering for primary prevention of coronary heart disease.

When used for the secondary prevention of coronary heart disease, treatment with an inhibitor of hydroxymethylglutaryl-coenzyme-A reductase results in worthwhile benefit that clearly exceeds any risk in patients whose risk of coronary death is 1.5% or more per year. This evidence can be extrapolated logically to primary prevention of coronary disease provided that treatment is targeted at those with similar or higher risk. We present a table that refines previously proposed methods of risk prediction. The table identifies subjects who have the specified degree of coronary risk; shows the serum cholesterol concentration that confers that degree or risk in the individual; and identifies subjects who will not have this degree of risk, irrespective of their cholesterol concentration. It is simple enough for use in ordinary practice. The table highlights the predominant effect of age on coronary risk; a person who is free of vascular disease and younger than 52 years is unlikely to have the specified degree of risk. Even in older people (60-70 years) several risk factors are generally required to attain this degree of risk. Some people are candidates for lipid- lowering drug treatment with serum cholesterol as low as 5.5 mmol/L, whereas others with cholesterol as high as 9.0 mmol/L are not. Although cholesterol lowering is a powerful method for preventing coronary events in people at high risk, cholesterol measurement by itself is not a good way to identify those with high coronary risk. The method can be adapted readily to target a different level of coronary risk as new evidence on the benefit and risk of treatment becomes available.

Non-pharmacological therapy of hypertension.

Weight reduction, moderate sodium restriction and alcohol reduction all lower blood pressure significantly in the short-term, and appear feasible in the long-term. Dynamic exercise may have a useful role in selected patients. Cessation of cigarette smoking has no important effect on blood pressure itself but is likely to improve the prognosis. No other non-pharmacological intervention warrants a place in routine management on present evidence. Regimens involving combined reduction in weight, salt and alcohol have proved less effective than drug therapy in terms of blood pressure reduction. Hypertensive patients may be shifted from just above some arbitrary intervention level to just below it by non-pharmacological treatment, and the perceived benefits of non-pharmacological management may be offset by an increased risk of vascular complications related to suboptimal blood pressure control. Moreover even simple measures such as moderate sodium restriction may affect some aspects of quality of life adversely. Non-pharmacological measures should generally be regarded as useful adjuncts to antihypertensive drug therapy rather than alternatives to it.

Investigation of selected patients with hypertension by the rapid-sequence intravenous urogram.

The rapid-sequence intravenous urogram (IVU) has tended to fall from favour for investigating hypertension because of its perceived imprecision for detecting renovascular disease. However, no study has examined the value of the IVU as a screening test in appropriately selected patients. We have analysed the diagnostic yield of the rapid-sequence IVU in hypertensive patients selected for features suggesting renal or renovascular disease in a retrospective review of case records from a hypertension clinic. The IVU was abnormal in 27% (95% CI 21-32%) of 241 consecutive patients. The most common abnormalities were chronic pyelonephritis (6%); proven renovascular disease (5%); stone (4%); possible renovascular disease and simple cyst (each 3%); hydronephrosis (2%); and tumour and active tuberculosis (each 1%). The IVU led to intervention aiming to correct hypertension in 5% (95% CI 2-8%) of patients, and revealed an abnormality needing intervention in its own right in 4% (95% CI 2-6%). The IVU led to unnecessary invasive investigation in 3% of cases. Individual abnormalities could not be predicted from the clinical or laboratory features. The initial investigation in hypertensive patients with suspected renal or renovascular disease should be a general purpose test able to detect a wide range of abnormalities. The rapid-sequence IVU is the only single test capable of satisfying this requirement. In patients with features suggesting renovascular disease, a normal rapid-sequence IVU excludes renovascular disease with 93% probability, but is an imperfect screening test since it fails to diagnose about 20% of cases. Renal arteriography should be done despite a normal IVU when it is essential to exclude renovascular disease.

Ultrasound scanning and 99mTc sulphur colloid scintigraphy in diagnosis of Budd-Chiari syndrome.

Ultrasound scanning and 99mTc sulphur colloid scintigraphy are widely used in the diagnosis of the Budd-Chiari syndrome and have been compared at the time of presentation in 18 patients in whom the diagnosis was subsequently confirmed by histology and hepatic venography. Ultrasound was diagnostic in 16 (87%). The findings seen most often included hepatic vein abnormalities, caudate lobe hypertrophy with decreased reflectivity and compression of the inferior vena cava. Additional information not shown by scintigraphy included intracaval tumour, or thrombosis, and concomitant portal vein thrombosis. Although scintigraphic abnormalities were present in all patients, only in three (17%) was the 'classical' appearance of increased uptake and/or enlargement of the caudate lobe present. In one patient with nonspecific abnormalities on ultrasound, scintigraphy gave a positive diagnosis and it is in such cases that scintigraphy should continue to be used.

Budd-Chiari syndrome presenting as fulminant hepatic failure.

Two cases of the Budd-Chiari syndrome are described in whom the diagnosis was finally confirmed at necropsy. The presentation was with encephalopathy, occurring within eight weeks of first symptoms and coming therefore within the definition of fulminant hepatic failure. In one, thought to have non-A, non-B hepatitis, encephalopathy progressed to grade 4 coma with death 12 days after presentation. In the other, mistakenly thought to have intra-abdominal malignancy, an exploratory laparotomy exacerbated the encephalopathy with death three weeks later. In neither case were non-invasive investigations, such as ultrasound and isotope scanning, carried out which might have facilitated an earlier diagnosis and consideration for orthotopic liver transplantation, probably the most appropriate form of therapy for these very severe cases.

Fluorogenic substrate detection of viable intracellular and extracellular pathogenic protozoa.

Viable Leishmania promastigotes and amastigotes were detected by epifluorescence microscopy with fluorescein diacetate being used to mark living parasites and the nucleic acid-binding compound ethidium bromide to stain dead cells. This procedure is superior to other assays because it is faster and detects viable intracellular as well as extracellular Leishmania. Furthermore, destruction of intracellular pathogens by macrophages is more accurately determined with fluorescein diacetate than with other stains. The procedure may have applications in programs to develop drugs and vaccines against protozoa responsible for human and animal disease.

Altered plasma drug binding in cancer: role of alpha 1-acid glycoprotein and albumin.

Altered concentrations of serum proteins often accompany malignant disease. The effect of these changes on drug binding was studied with lidocaine, a basic drug, and tolbutamide, an acidic drug. Patients with cancer had increased serum concentrations of the acute-phase protein alpha 1-acid glycoprotein (AAG) and lowered serum concentration of albumin. In association with these changes lidocaine binding was increased at all concentrations studied (predialysis concentrations 2, 6, and 10 microgram . ml-1) and that of tolbutamide was decreased at the highest concentration (200 microgram . ml-1). Not all of the increase in lidocaine binding was explicable on the basis of increased serum AAG concentration. Estimation of binding parameters with a model with two independent sites showed increased affinity at the high affinity site in cancer patients with no change in the calculated number of binding sites. Therefore, in cancer there is increased lidocaine binding in association with increased AAG concentrations. We also record the novel observation of a change in the intrinsic properties of the high affinity binding site.

Intestinal bacterial metabolism of protein and bile acids: role in pathogenesis of hepatic disease after jejuno-ileal bypass surgery.

Jejunal bacterial colonization and intestinal metabolism of bile acids and protein by bacteria have been investigated in 12 patients with abnormal liver histology following jeujno-ileal bypass surgery for obesity. Aerobic and/or anaerobic colonic flora was present in jejunal aspirates from 8 of 12 bypass patients, but in none of the controls. Intestinal protein metabolism and bile acid deconjugation (measured by urinary indican excretion and 14C-glycocholic acid breath test) was significantly enhanced in bypass patients. Intestinal bacterial overgrowth, with abnormal intestinal metabolism by bacteria of ingested nutrients and bile acids, could contribute to hepatic disease after bypass surgery via the production of endogenous hepatotoxins.