Genetic variation in aldosterone synthase predicts plasma glucose levels.

The mineralocorticoid hormone, aldosterone, is known to play a role in sodium homeostasis. We serendipitously found, however, highly significant association between single-nucleotide polymorphisms in the aldosterone synthase gene and plasma glucose levels in a large population of Chinese and Japanese origin. Two polymorphisms--one in the putative promoter (T-344C) and another resulting in a lysine/arginine substitution at amino acid 173, which are in complete linkage disequilibrium in this population--were associated with fasting plasma glucose levels (P = 0.000017) and those 60 (P = 0.017) and 120 (P = 0.0019) min after an oral glucose challenge. A C/T variant in intron 1, between these polymorphisms, was not associated with glucose levels. Arg-173 and -344C homozygotes were most likely to be diabetic [odds ratio 2.51; 95% confidence interval (C.I.) 1.39-3.92; P = 0.0015] and have impaired fasting glucose levels (odds ratio 3.53; 95% C.I. 2.02-5.5; P = 0.0000036). These results suggest a new role for aldosterone in glucose homeostasis.

Intracellular reduction of selenite into glutathione peroxidase. Evidence for involvement of NADPH and not glutathione as the reductant.

Selenium (Se) in selenite is present in an oxidized state, and must be reduced for it to be incorporated as selenocysteine into selenoenzymes such as glutathione peroxidase (GPx). In vitro, Se, as in selenite, can be reduced utilizing glutathione (GSH) and glutathione reductase (GRed). We determined the effects of decreasing GSH levels, inhibiting GRed activity, and decreasing cellular NADPH on the selenite-dependent rate of GPx synthesis in cultured cells: PC3, CHO, and the E89 glucose-6-phosphate dehydrogenase (G-6-PD)-deficient cell line. A novel statistical analysis method was developed (using Box Cox transformed regression and a bootstrap method) in order to assess the effects of these manipulations singly and in combinations. Buthionine sulfoximine (BSO) was used to decrease GSH levels, 1,3 bis-(2 chloroethyl)-1 -nitrosourea (BCNU) was used to inhibit GRed activity and methylene blue (MB) was used to decrease cellular NADPH levels. This statistical method evaluates the effects of BSO, BCNU, MB and selenite alone and in combinations on GPx activity. Decreasing the GSH level (< 5% of control) did not have an effect on the selenite-dependent rate of GPx synthesis in PC3 or CHO cells, but did have a small inhibitory effect on the rate of GPx synthesis in E89 cells. Inhibiting GRed activity was also associated with either no effect (CHO, E89) or a small effect (PC3) on GPx activity. In contrast, decreasing NADPH levels in cells treated with MB was associated with a large decrease in the selenite-dependent rate of GPx synthesis to 36, 34 and 25% of control in PC3, CHO, and E89 cells, respectively. The effects of BSO plus BCNU were not synergistic in any of the cell lines. The effects of BSO plus MB were synergistic in G-6-PD-deficient E89 cells, but not in PC3 or CHO cells. We therefore conclude that under normal culture conditions, NADPH, and not glutathione, is the primary reductant of Se in selenite to forms that are eventually incorporated into GPx. For cells with abnormal ability to generate NADPH, lowering the GSH levels had a small effect on selenite-dependent GPx synthesis. GRed activity is not required for the selenite-dependent synthesis of GPx.

Clustering and the design of preference-assessment surveys in healthcare.

OBJECTIVE: To show cluster analysis as a potentially useful tool in defining common outcomes empirically and in facilitating the assessment of preferences for health states. DATA SOURCES: A survey of 224 patients with ventricular arrhythmias treated at Kaiser Permanente of Northern California. STUDY DESIGN/METHODS: Physical functioning was measured using the Duke Activity Status Index (DASI), and mental status and vitality using the Medical Outcomes Study Short Form-36 items (SF-36). A "k-means" clustering algorithm was used to identify prototypical health states, in which patients in the same cluster shared similar responses to items in the survey. PRINCIPAL FINDINGS: The clustering algorithm yielded four prototypical health states. Cluster 1 (21 percent of patients) was characterized by high scores on physical functioning, vitality, and mental health. Cluster 2 (33 percent of patients) had low physical function but high scores on vitality and mental health. Cluster 3 (29 percent of patients) had low physical function and low vitality but preserved mental health. Cluster 4 (17 percent of patients) had low scores on all scales. These clusters served as the basis of written descriptions of the health states. CONCLUSIONS: Employing a clustering algorithm to analyze health status survey data enables researchers to gain a data-driven, concise summary of the experiences of patients.

PAH extraction and estimation of plasma flow in human postischemic acute renal failure.

We determined the effect of postischemic injury to the human renal allograft on p-aminohippurate (PAH) extraction (E(PAH)) and renal blood flow. We evaluated renal function in 44 allograft recipients on two occasions: 1-3 h after reperfusion (day 0) and again on postoperative day 7. On day 0 subsets underwent intraoperative determination of renal blood flow (n = 35) by Doppler flow meter and E(PAH) (n = 25) by renal venous assay. Blood flow was also determined in another subset of 16 recipients on postoperative day 7 by phase contrast-cine-magnetic resonance imaging, and E(PAH) was computed from the simultaneous PAH clearance. Glomerular filtration rate (GFR) on day 7 was used to divide subjects into recovering (n = 23) and sustained (n = 21) acute renal failure (ARF) groups, respectively. Despite profound depression of GFR in the sustained ARF group, renal plasma flow was only slightly depressed, averaging 296 +/- 162 ml. min(-1). 1.73 m(-2) on day 0 and 202 +/- 72 ml. min(-1). 1.73 m(-2) on day 7, respectively. These values did not differ from corresponding values in the recovering ARF group: 252 +/- 133 and 280 +/- 109 ml. min(-1). 1.73 m(-2), respectively. E(PAH) was profoundly depressed on day 0, averaging 18 +/- 14 and 10 +/- 7% in recovering and sustained ARF groups, respectively, vs. 86 +/- 6% in normal controls (P < 0.001). Corresponding values on day 7 remained significantly depressed at 65 +/- 20 and 11 +/- 22%, respectively. We conclude that postischemic injury to the renal allograft results in profound impairment of E(PAH) that persists for at least 7 days, even after the onset of recovery. An ensuing reduction in urinary PAH clearance results in a gross underestimate of renal plasma flow, which is close to the normal range in the initiation, maintenance, and recovery stages of this injury.

Empirically defined health states for depression from the SF-12.

OBJECTIVE: To define objectively and describe a set of clinically relevant health states that encompass the typical effects of depression on quality of life in an actual patient population. Our model was designed to facilitate the elicitation of patients' and the public's values (utilities) for outcomes of depression. DATA SOURCES: From the depression panel of the Medical Outcomes Study. Data include scores on the 12-Item Short Form Health Survey (SF-12) as well as independently obtained diagnoses of depression for 716 patients. Follow-up information, one year after baseline, was available for 166 of these patients. METHODOLOGY: We use k-means cluster analysis to group the patients according to appropriate dimensions of health derived from the SF-12 scores. Chi-squared and exact permutation tests are used to validate the health states thus obtained, by checking for baseline and longitudinal correlation of cluster membership and clinical diagnosis. PRINCIPAL FINDINGS: We find, on the basis of a combination of statistical and clinical criteria, that six states are optimal for summarizing the range of health experienced by depressed patients. Each state is described in terms of a subject who is typical in a sense that is articulated with our cluster-analytic approach. In all of our models, the relationship between health state membership and clinical diagnosis is highly statistically significant. The models are also sensitive to changes in patients' clinical status over time. CONCLUSIONS: Cluster analysis is demonstrably a powerful methodology for forming clinically valid health states from health status data. The states produced are suitable for the experimental elicitation of preference and analyses of costs and utilities.

Angiotensin-converting enzyme inhibitor for slowing progression of diabetic and nondiabetic kidney disease.

A growing body of evidence suggests that agents that inhibit the angiotensin-converting enzyme are renoprotective. In experimental animal models of chronic renal injury, such renoprotection can virtually eliminate progression of the renal injury, provided that therapy is started at the time of injury. In humans with chronic renal injury, renoprotection has been successfully demonstrated only late in the course of the renal disease. The rate of progression to end-stage renal failure can be delayed, but progression continues at a slower pace. Further study is required to determine whether earlier intervention can better preserve nephron structure and function. A strategy for future trials is recommended. It emphasizes more sensitive outcome measures so as to achieve greater statistical power.

Modeling of progressive glomerular injury in humans with lupus nephritis.

We studied glomerular function longitudinally for 36-120 mo in 21 patients undergoing treatment for diffuse, proliferative lupus nephritis. We determined glomerular filtration rate (GFR) and glomerular oncotic pressure (IIGC) and computed the two-kidney ultrafiltration coefficient (Kf) at 6- to 12-mo intervals. The relationships and cross talk among the three variables over time were then analyzed by eigenfunction regression and canonical correlations. We also performed a morphometric analysis of serial biopsies and computed single-nephron Kf in patent glomeruli at baseline and after 36-94 mo of follow-up. Patients were divisible into progressors (n = 12) or nonprogressors (n = 9) according to the presence or absence, respectively, of an irrevocable decline in GFR over time. Examination of longitudinal variables revealed GFR to be strongly related to Kf in all patients and inversely related to IIGC in progressors. By serial morphometric analysis we observed a threefold increase in the prevalence of global sclerosis in progressors but unchanged prevalence in nonprogressors. Whereas single-nephron Kf of remnant glomeruli increased to supernormal levels in nonprogressors, the absence of this compensatory phenomenon in progressors permitted GFR and Kf to decline in parallel with the declining number of functional glomeruli.

Covariability of V3 loop amino acids.

We reanalyzed for covariability a set of 308 human immunodeficiency virus type 1 (HIV-1) V3 loop amino acid sequences from the B envelope sequence subtype previously analyzed by Korber et al.,1 as well as a new set of 440 sequences that also included substantial numbers of sequences from subtypes A, D, and E. We used the measure employed by Korber et al., essentially the likelihood ratio statistic for independence, plus two additional measures as well as clade information to examine the new set and both data sets simultaneously. We set forth the following conclusions and observations. The eight most highly connected sites identified through these statistical approaches included all of the six residues previously shown to have determining roles in structure, immunologic recognition, virus phenotype, and host range; each of the seven pairs of covariant sites found by Korber were signaled by our additional two measures in the set of 308 sequences, although 2 or 3 dropped out of the examination of the set of 440 when the requirement of stringent significance was applied for some or all of the three tests, respectively; using the same criteria, a total of 20 (including 5 Korber et al. pairs) or a total of 6 (including 4 Korber et al. pairs) were found when the set of 440 was added. Several limitations to statistical analysis of this type of HIV sequence data were also noted. For example, the data sets were, by historical necessity, collected haphazardly. For example, it was not possible to separate substantially sized groups out according to time of or since infection, disease status, antiviral treatment, geography, etc. There was also an enormous "wealth of significance" within the data. For example, for one measure the 440 data set showed 233 of the 465 pairs of sites with a likelihood ratio statistic of < 0.001. Last, most sites had consensus amino acids in 80% or more of the sequences; hence, there was an absence of data on many combinations of amino acids. Given the observed linkage between sites shown to be covariable and those known to have critical biological function, the statistical approaches we and Korber et al. have outlined may find use in predicting critical structural features of HIV proteins as targets for therapeutic intervention.

Bayes risk weighted vector quantization with posterior estimation for image compression and classification.

Classification and compression play important roles in communicating digital information. Their combination is useful in many applications, including the detection of abnormalities in compressed medical images. In view of the similarities of compression and low-level classification, it is not surprising that there are many similar methods for their design. Because some of these methods are useful for designing vector quantizers, it seems natural that vector quantization (VQ) is explored for the combined goal. We investigate several VQ-based algorithms that seek to minimize both the distortion of compressed images and errors in classifying their pixel blocks. These algorithms are investigated with both full search and tree-structured codes. We emphasize a nonparametric technique that minimizes both error measures simultaneously by incorporating a Bayes risk component into the distortion measure used for the design and encoding. We introduce a tree-structured posterior estimator to produce the class posterior probabilities required for the Bayes risk computation in this design. For two different image sources, we demonstrate that this system provides superior classification while maintaining compression close or superior to that of several other VQ-based designs, including Kohonen's (1992) "learning vector quantizer" and a sequential quantizer/classifier design.

Pruned tree-structured vector quantization of medical images with segmentation and improved prediction.

The authors use predictive pruned tree-structured vector quantization for the compression of medical images. Their goal is to obtain a high compression ratio without impairing the image quality, at least so far as diagnostic purposes are concerned. The authors use a priori knowledge of the class of images to be encoded to help them segment the images and thereby to reserve bits for diagnostically relevant areas. Moreover, the authors improve the quality of prediction and encoding in two additional ways: by increasing the memory of the predictor itself and by using ridge regression for prediction. The improved encoding scheme was tested via computer simulations on a set of mediastinal CT scans; results are compared with those obtained using a more conventional scheme proposed recently in the literature. There were remarkable improvements in both the prediction accuracy and the encoding quality, above and beyond what comes from the segmentation. Test images were encoded at 0.5 bit per pixel and less without any visible degradation for the diagnostically relevant region.

Effect of treatment for Hodgkin's disease on pulmonary function: results of a prospective study.

PURPOSE: Because each of very different treatments for Hodgkin's disease (HD) may result in a high rate of cure, attention is currently focused on toxicity. This prospective study was designed to assess the effects of mediastinal irradiation and bleomycin chemotherapy on pulmonary function. PATIENTS AND METHODS: Patients were treated from 1980 to 1990 on randomized controlled trials at Stanford University. Pulmonary function was tested before treatment (baseline), early after treatment (< 15 months), and more than 36 months posttherapy. Treatment options in the 145 patients were grouped as I (mediastinal radiotherapy), II (mediastinal radiotherapy plus bleomycin), and III (bleomycin) for analyses of variance (ANOVAs). A variety of regression models were used to predict early and late effects on pulmonary function. RESULTS: A decrease in forced vital capacity (FVC) and diffusing capacity (DLCO) in the first 15 months after treatment followed by recovery after 36 months was observed for most patients. Patients who received mediastinal radiotherapy (RT) had a more pronounced reduction in pulmonary function and less complete recovery. Overall, 3 or more years after treatment, 32% of group I patients, 37% of group II patients, and 19% of group III patients had FVC values less than 80% of predicted, while only 7% of patients had a DLCO less than 80% of predicted. Linear regression identified baseline measurement as the only significant predictor of change in percent predicted FVC or DLCO; patients with higher baseline values had greater decrements after therapy. Mantle RT was the only significant treatment variable, predictive of FVC and DLCO within 15 months and FVC at 36 or more months. No patient experienced pulmonary toxicity severe enough to require hospitalization. CONCLUSION: This prospective analysis of pulmonary function after treatment for HD showed that mediastinal RT was the only treatment variable that achieved statistical significance. Although there were no significant interactions between mediastinal RT and bleomycin or Adriamycin (doxorubicin; Adria Laboratories, Columbus, OH) chemotherapy, the patient numbers were small after correction for mediastinal mass size and drug regimen such that an effect could have been missed. The mild reduction in pulmonary function should be factored into the overall assessment of morbidity risk for each of the potentially curative treatments included in this study. As with all reports of late effects, these data should be interpreted with respect to the population tested, details of the treatment administered, methods of measurement, and length of follow-up.

Thoracic CT images: effect of lossy image compression on diagnostic accuracy.

PURPOSE: To evaluate the effects of lossy image (noninvertible) compression on diagnostic accuracy of thoracic computed tomographic images. MATERIALS AND METHODS: Sixty images from patients with mediastinal adenopathy and pulmonary nodules were compressed to six different levels with tree-structured vector quantization. Three radiologists then used the original and compressed images for diagnosis. Unlike many previous receiver operating characteristic-based studies that used confidence rankings and binary detection tasks, this study examined the sensitivity and predictive value positive scores from nonbinary detection tasks. RESULTS: At the 5% significance level, there was no statistically significant difference in diagnostic accuracy of image assessment at compression rates of up to 9:1. CONCLUSION: The techniques presented for evaluation of image quality do not depend on the specific compression algorithm and provide a useful approach to evaluation of the benefits of any lossy image processing technique.

Relationships among protein and albumin concentrations and oncotic pressure in nephrotic plasma.

We determined oncotic pressure (pi) by membrane osmometry and assayed total protein (TP) and albumin (Alb) concentrations in plasma of 102 nephrotic subjects and 27 healthy controls. All three quantities were markedly depressed in the nephrotic group. When plasma was serially diluted and concentrated, nephrotic but not control plasma also exhibited a highly variable change point in the nonlinear relationship between TP or Alb and pi. Absent a unique change point, we developed quadratic models which incorporated TP, Alb, and (TP x Alb) to prospectively predict pi in unperturbed plasma. The ability of the most successful quadratic model to predict pi in afferent or efferent arteriolar plasma was limited; the prediction errors reached 10 mmHg in nephrotic and 6 mmHg in control subjects. The nephrotic model coefficients also differed significantly from control and pointed to an important influence of nonalbumin proteins on pi in nephrotic plasma. Investigation of the intrinsic membrane properties of diseased glomerular capillary walls requires precise knowledge of pi. For this purpose we recommend that pi be directly determined by membrane osmometry rather than calculated from protein concentration(s).

Tree-structured vector quantization of CT chest scans: image quality and diagnostic accuracy.

The authors apply a lossy compression algorithm to medical images, and quantify the quality of the images by the diagnostic performance of radiologists, as well as by traditional signal-to-noise ratios and subjective ratings. The authors' study is unlike previous studies of the effects of lossy compression in that they consider nonbinary detection tasks, simulate actual diagnostic practice instead of using paired tests or confidence rankings, use statistical methods that are more appropriate for nonbinary clinical data than are the popular receiver operating characteristic curves, and use low-complexity predictive tree-structured vector quantization for compression rather than DCT-based transform codes combined with entropy coding. The authors' diagnostic tasks are the identification of nodules (tumors) in the lungs and lymphadenopathy in the mediastinum from computerized tomography (CT) chest scans. Radiologists read both uncompressed and lossy compressed versions of images. For the image modality, compression algorithm, and diagnostic tasks the authors consider, the original 12 bit per pixel (bpp) CT image can be compressed to between 1 bpp and 2 bpp with no significant changes in diagnostic accuracy. The techniques presented here for evaluating image quality do not depend on the specific compression algorithm and are useful new methods for evaluating the benefits of any lossy image processing technique.

Probability estimation for biomedical classification problems.

Suppose that we wish to know the probability that an object belongs to a class. For example, we may wish to estimate the probability that a patient has a particular disease, given a set of symptoms, or we may wish to know the probability that a novel peptide binds to a receptor, given the peptide's amino-acid composition. The conventional approach is to first use a classification algorithm to find partitions in feature space and to assign each partition to a class, and then to estimate the conditional probabilities as the proportion of patients or peptides that are correctly and incorrectly classified in each partition. Unfortunately, this estimation method often gives probability estimates that are in error by 20% or more, and thus can cause incorrect decisions. We have implemented and compared alternative methods. In Monte Carlo simulations the alternative methods are substantially more accurate than is the current method.

PAH extraction and estimation of plasma flow in diseased human kidneys.

We have analyzed the efficiency with which p-amino-hippuric acid (PAH) is extracted (EPAH) by patients with healthy kidneys (n = 13) or kidneys damaged by chronic cyclosporin nephropathy (n = 21) or primary glomerulopathy (n = 12); respective values (mean +/- SE) for EPAH were 0.87 +/- 0.03, 0.77 +/- 0.03, and 0.69 +/- 0.04. Judged by a 131I-hippuran-to-PAH clearance ratio of 0.75 +/- 0.05, extraction ratio of hippuran was less efficient than EPAH in three glomerulopathic patients. A direct relationship was defined between EPAH and glomerular filtration rate (GFR) (r = 0.54) or calculated efferent oncotic pressure (IIE; r = 0.41, P less than 0.01). Curve fitting by means of quadratic spline functions revealed GFR and IIE to be additive in predicting EPAH (R2 = 0.45). Linear model prediction methods and a sample reuse technique failed to predict EPAH reliably from GFR and preglomerular oncotic pressure (IIA); however, 95% prediction intervals exceed 0.30 EPAH units in width. We conclude that oncotic pressure (presumably reflecting albumin concentration) along with GFR is predictive of EPAH depression in humans with chronic renal disease. However, even sophisticated curve-fitting techniques are too imprecise for accurate prediction of EPAH in a given individual. We submit that renal venous sampling to determine EPAH continues to be necessary for the accurate determination of the rate of plasma flow in the injured human kidney.

Predicting chemically induced duodenal ulcer and adrenal necrosis with classification trees.

Binary tree-structured statistical classification algorithms and properties of 56 model alkyl nucleophiles were brought to bear on two problems of experimental pharmacology and toxicology. Each rat of a learning sample of 745 was administered one compound and autopsied to determine the presence of duodenal ulcer or adrenal hemorrhagic necrosis. The cited statistical classification schemes were then applied to these outcomes and 67 features of the compounds to ascertain those characteristics that are associated with biologic activity. For predicting duodenal ulceration, dipole moment, melting point, and solubility in octanol are particularly important, while for predicting adrenal necrosis, important features include the number of sulfhydryl groups and double bonds. These methods may constitute inexpensive but powerful ways to screen untested compounds for possible organ-specific toxicity. Mechanisms for the etiology and pathogenesis of the duodenal and adrenal lesions are suggested, as are additional avenues for drug design.

Prognostic significance of actual dose intensity in diffuse large-cell lymphoma: results of a tree-structured survival analysis.

While diffuse large-cell lymphoma (DLCL) is considered to be highly curable with current therapy, treatment failures are observed even with intensive combination chemotherapy regimens. In order to study the prognostic significance of actual dose intensity of chemotherapy in DLCL, we retrospectively analyzed 115 previously untreated patients treated as Stanford between 1975 and 1986 with cyclophosphamide, Adriamycin (doxorubicin; Adria Laboratories, Columbus, OH), vincristine, and prednisone (CHOP), methotrexate, bleomycin, Adriamycin, cyclophosphamide, vincristine, and dexamethasone ([M]BACOD), or methotrexate, Adriamycin, cyclosphosphamide, vincristine, prednisone, and bleomycin (MACOP-B). The actual relative dose intensity (RDI), the amount of drug actually administered to each patient during the first 12 weeks of therapy, was calculated as standardized to CHOP and analyzed in addition to clinical factors prognostic for survival by univariate analysis. Multivariate recursive partitioning (tree-structured) survival analysis identified the actual RDI of Adriamycin greater than 75% as the single most important predictor of survival. A model incorporating the actual RDI of Adriamycin and performance status, in combination with serum lactate dehydrogenase (LDH) and extranodal disease, defined three overall prognostic groups of patients with respective 3-year survival rates of 89%, 63%, and 18%. The three prognostic groups remained distinct, even when restricted to complete responders. This model was also predictive of survival when dose intensity was analyzed relative to the optimum dose defined for each of the three regimens and when applied to a subgroup of patients aged 50 years or younger. We conclude that actual RDI is an important prognostic factor for survival in DLCL and that analysis of RDI early in the course of treatment may allow modification of the treatment plan.

Predicting 1-year outcome following acute myocardial infarction: physicians versus computers.

Whether decision rules derived statistically from patient data can produce better decisions than an expert clinician or a model of the expert clinician (expert system) is controversial. We examined this issue in the context of predicting cardiac death by 1 year for patients discharged from the hospital following acute myocardial infarction. Decision rules were derived from a base sample of 781 patients. These decision rules and three experienced cardiologists then estimated probability of death by 1 year for each patient in a separate test sample (n = 400). In our evaluation of the performance of the decision rules and physicians, we detected no differences, although the decision rules and physicians tended to classify the patients somewhat differently. Further multivariate analyses on the physicians' predictions showed that two of the physicians paid attention to somewhat different variables than the third physician. Lack of agreement among expert cardiologists would complicate modeling of a consensual decision-making process within the framework of an expert system.

Tree-structured survival analysis.

In this note, tree-structured recursive partitioning schemes for classification, probability class estimation, and regression are adapted to cover censored survival analysis. The only assumptions required are those which guarantee identifiability of conditional distributions of lifetime given covariates. Thus, the techniques are applicable to more general situations than are those of the famous semi-parametric model of Cox.