Emil Pitkin

Peeking Inside the Black Box: Visualizing Statistical Learning With Plots of Individual Conditional Expectation

This article presents individual conditional expectation (ICE) plots, a tool for visualizing the model estimated by any supervised learning algorithm. Classical partial dependence plots (PDPs) help visualize the average partial relationship between the predicted response and one or more features. In the presence of substantial interaction effects, the partial response relationship can be heterogeneous. Thus, an average curve, such as the PDP, can obfuscate the complexity of the modeled relationship. Accordingly, ICE plots refine the PDP by graphing the functional relationship between the predicted response and the feature for individual observations. Specifically, ICE plots highlight the variation in the fitted values across the range of a covariate, suggesting where and to what extent heterogeneities might exist. In addition to providing a plotting suite for exploratory analysis, we include a visual test for additive structure in the data-generating model. Through simulated examples and real datasets, we demonstrate how ICE plots can shed light on estimated models in ways PDPs cannot. Procedures outlined are available in the R package ICEbox.

Effective immunotherapy of rat glioblastoma with prolonged intratumoral delivery of exogenous heat shock protein Hsp70

Chaperone Hsp70 can activate adaptive immunity suggesting its possible application as an antitumor vaccine. To assess the therapeutic capacity of Hsp70 we administered purified chaperone into a C6 glioblastoma brain tumor and explored the viability and tumor size as well as interferon gamma (IFNγ) production and cytotoxicity of lymphocytes in the treated animals. Targeted intratumoral injection of Hsp70 resulted in its distribution within the area of glioblastoma, and caused significant inhibition of tumor progression as confirmed by magnetic resonance imaging. The delay in tumor growth corresponded to the prolonged survival of tumor‐bearing animals of up to 31 days versus 20 days in control. Continuous administration of Hsp70 with an osmotic pump increased survival even further (39 days). Therapeutic efficacy was associated with infiltration to glioblastoma of NK cells (Ly‐6c+) and T lymphocytes (CD3+, CD4+ and CD8+) as well as with an increase in the activity of NK cells (granzyme B production) and CD8+ T lymphocytes as shown by IFNγ ELISPOT assay. Furthermore, we found that Hsp70 treatment caused concomitantly, with a tenfold elevated IFNγ production, an increase in anti‐C6 tumor cytotoxicity of lymphocytes. In conclusion, continuous intratumoral delivery of Hsp70 demonstrates high therapeutic potential and therefore could be applied in the treatment of glioblastoma.

Recombinant interleukin-1 receptor antagonist conjugated to superparamagnetic iron oxide nanoparticles for theranostic targeting of experimental glioblastoma.

Cerebral edema commonly accompanies brain tumors and contributes to neurologic symptoms. The role of the interleukin-1 receptor antagonist conjugated to superparamagnetic iron oxide nanoparticles (SPION–IL-1Ra) was assessed to analyze its anti-edemal effect and its possible application as a negative contrast enhancing agent for magnetic resonance imaging (MRI). Rats with intracranial C6 glioma were intravenously administered at various concentrations of IL-1Ra or SPION–IL-1Ra. Brain peritumoral edema following treatment with receptor antagonist was assessed with high-field MRI. IL-1Ra administered at later stages of tumor progression significantly reduced peritumoral edema (as measured by MRI) and prolonged two-fold the life span of comorbid animals in a dose-dependent manner in comparison to control and corticosteroid-treated animals (P < .001). Synthesized SPION–IL-1Ra conjugates had the properties of negative contrast agent with high coefficients of relaxation efficiency. In vitro studies of SPION–IL-1Ra nanoparticles demonstrated high intracellular incorporation and absence of toxic influence on C6 cells and lymphocyte viability and proliferation. Retention of the nanoparticles in the tumor resulted in enhanced hypotensive T2-weighted images of glioma, proving the application of the conjugates as negative magnetic resonance contrast agents. Moreover, nanoparticles reduced the peritumoral edema confirming the therapeutic potency of synthesized conjugates. SPION–IL-1Ra nanoparticles have an anti-edemal effect when administered through a clinically relevant route in animals with glioma. The SPION–IL-1Ra could be a candidate for theranostic approach in neuro-oncology both for diagnosis of brain tumors and management of peritumoral edema.

70-kDa heat shock protein coated magnetic nanocarriers as a nanovaccine for induction of anti-tumor immune response in experimental glioma

Nanovaccines based on superparamagnetic iron oxide nanoparticles (SPIONs) provide a novel approach to induce the humoral and cell-based immune system to fight cancer. Herein, we increased the immunostimulatory capacity of SPIONs by coating them with recombinant heat shock protein 70 (Hsp70) which is known to chaperone antigenic peptides. After binding, Hsp70-SPIONs deliver immunogenic peptides from tumor lysates to dendritiс cells (DCs) and thus stimulate a tumor-specific, CD8+ cytotoxic T cell response. We could show that binding activity of Hsp70-SPIONs to the substrate-binding domain (SBD) is highly dependent on the ATPase activity of its nucleotide-binding domain NBD), as shown by (31)P NMR spectroscopy. Immunization of C6 glioma-bearing rats with DCs pulsed with Hsp70-SPIONs and tumor lysates resulted in a delayed tumor progression (as measured by MRI) and an increased overall survival. In parallel an increased IFNγ secretion were detected in the serum of these animals and immunohistological analysis of subsequent cryosections of the glioma revealed an enhanced infiltration of memory CD45RO+ and cytotoxic CD8+ T cells. Taken together the study demonstrates that magnetic nanocarriers such as SPIONs coated with Hsp70 can be applied as a platform for boosting anti-cancer immune responses.

Two‐stage implantation of the skin‐ and bone‐integrated pylon seeded with autologous fibroblasts induced into osteoblast differentiation for direct skeletal attachment of limb prostheses

Angio- and osteogenesis following the two-stage (TS) implantation of the skin- and bone-integrated pylon seeded with autologous fibroblasts was evaluated. Two consecutive animal substudies were undertaken: intramedullary subcutaneous implantation (15 rabbits) and a TS transcutaneous implantation (12 rabbits). We observed enhanced osseointegrative properties of the intramedullary porous component seeded with fibroblasts induced into osteoblast differentiation, as compared to the untreated porous titanium pylon. The three-phase scintigraphy and subsequent histological analysis showed that the level of osteogenesis was 1.5-fold higher than in the control group, and significantly so (p < 0.05). The biocompatibility was further proved by the absence of inflammatory response or encapsulation and sequestration on the histology assay. Treatment of the transcutaneous component with autologous fibroblasts was associated with nearly a 2-fold decrease in the period required for the ingrowth of dermal and subdermal soft tissues into the implant surface, as compared to the untreated porous titanium component. Direct dermal attachment to the transcutaneous implant prevented superficial and deep periprosthetic infections in rabbits in vivo.

Covariance Adjustments for the Analysis of Randomized Field Experiments

Background: It has become common practice to analyze randomized experiments using linear regression with covariates. Improved precision of treatment effect estimates is the usual motivation. In a series of important articles, David Freedman showed that this approach can be badly flawed. Recent work by Winston Lin offers partial remedies, but important problems remain. Results: In this article, we address those problems through a reformulation of the Neyman causal model. We provide a practical estimator and valid standard errors for the average treatment effect. Proper generalizations to well-defined populations can follow. Conclusion: In most applications, the use of covariates to improve precision is not worth the trouble.

Minimally Invasive Port Access Approach for Reoperations on the Mitral Valve

BACKGROUND In patients requiring a second-time or more operation on the mitral valve (MV), we assessed whether the outcomes of the minimally invasive port access approach (port access group) were equivalent to those of the traditional redo sternotomy approach (redo sternotomy group). METHODS In a retrospective review (1998-2011), 409 patients had previous MV operations requiring a second-time or more MV reintervention. Of those, 67 patients had the port access approach, and 342 had the redo sternotomy approach. Of the latter, 220 met the inclusion criteria because emergencies, patients with endocarditis, and those requiring concomitant procedures involving aortic valve and aorta were excluded. RESULTS New York Heart Association class 2 or above, age, atrial fibrillation, and surgical indications were similar in both groups. The port access group had more patients with previous MV repair (78% [n = 52] vs 41% [n = 90], p < 0.01) than with MV replacement (19% [n = 13) vs 53% [n = 116], p < 0.01). Concomitant procedures were similar (20% [n = 14] vs 27% [n = 59], p = 0.4). The MV re-repair rates were similar (19% [n = 10] vs 22% [n = 20], p = 1). The cardiopulmonary bypass times (153 ± 42 minutes vs 172 ± 83 minutes, p = 0.07) and aortic cross-clamping times (104 ± 38 minutes versus 130 ± 71 minutes, p < 0.01) were lower in the port access group. Mortality was lower in the port access group, although not significantly (3.0% [n = 2] vs 6.0% [n = 13], p = 0.5). The rates of postoperative stroke were similar (3.0% [n = 2] vs 3.2% [n = 7], p = 1). On postoperative echocardiography, freedom from mitral regurgitation >2+ was 100% in the port access group and 99% in the redo sternotomy group. The mean hospital length of stay was 11 ± 15 days versus 14 ± 12 days (p = 0.07). CONCLUSIONS The port access approach can be safely adopted for reoperations on the MV without compromising postoperative mortality or MV function.

Misspecified Mean Function Regression Making Good Use of Regression Models That Are Wrong

There are over three decades of largely unrebutted criticism of regression analysis as practiced in the social sciences. Yet, regression analysis broadly construed remains for many the method of choice for characterizing conditional relationships. One possible explanation is that the existing alternatives sometimes can be seen by researchers as unsatisfying. In this article, we provide a different formulation. We allow the regression model to be incorrect and consider what can be learned nevertheless. To this end, the search for a correct model is abandoned. We offer instead a rigorous way to learn from regression approximations. These approximations, not “the truth,” are the estimation targets. There exist estimators that are asymptotically unbiased and standard errors that are asymptotically correct even when there are important specification errors. Both can be obtained easily from popular statistical packages.

Detection of experimental myocardium infarction in rats by MRI using heat shock protein 70 conjugated superparamagnetic iron oxide nanoparticle

UNLABELLED Superparamagnetic iron-oxide based contrast agents can provide important diagnostic information regarding the assessment of cardiac inflammatory diseases. The aim of the study was to analyze whether nanoparticles conjugated to recombinant 70-kDa heat shock protein (Hsp70-SPION) can be applied for the detection of acute myocardium infarct by MRI. Cellular experiments demonstrated increased CD40-mediated uptake of Hsp70-SPIONs in comparison to non-conjugated SPIONs. Following induction of an acute infarct in rats by ligation of the left anterior descending artery SPIONs and Hsp70-SPION conjugates were injected intravenously on day 4. The animals underwent sequential MRI that showed the presence of the particles in the infarcted zone. Subsequent biodistribution analyses with the help of method on non-linear magnetic response indicated the preferential accumulation of the Hsp70-SPIONs in the heart tissue that was further confirmed with histological analyses. The study demonstrated that an acute infarct can be visualized by MRI using Hsp70-functionalized SPION conjugates. FROM THE CLINICAL EDITOR Superparamagnetic iron oxides nanoparticles (SPIONs) have been studied extensively as a contrast agent for MRI. Their tissue specificity can be further enhanced by conjugation with various ligands. In this study, the authors conjugated superparamagnetic nanoparticles to 70-kDa heat shock protein (Hsp70-SPION) to investigate the feasibility for the detection of acute myocardium infarct. The positive findings would suggest that this approach might be used clinically in the future.

What You Can Learn from Wrong Causal Models

It is common for social science researchers to provide estimates of causal effects from regression models imposed on observational data. The many problems with such work are well documented and widely known. The usual response is to claim, with little real evidence, that the causal model is close enough to the “truth” that sufficiently accurate causal effects can be estimated. In this chapter, a more circumspect approach is taken. We assume that the causal model is a substantial distance from the truth and then consider what can be learned nevertheless. To that end, we distinguish between how nature generated the data, a “true” model representing how this was accomplished, and a working model that is imposed on the data. The working model will typically be “wrong.” Nevertheless, unbiased or asymptotically unbiased estimates from parametric, semiparametric, and nonparametric working models can often be obtained in concert with appropriate statistical tests and confidence intervals. However, the estimates are not of the regression parameters typically assumed. Estimates of causal effects are not provided. Correlation is not causation. Nor is partial correlation, even when dressed up as regression coefficients. However, we argue that insights about causal effects do not require estimates of causal effects. We also discuss what can be learned when our alternative approach is not persuasive.