Kellie J. Archer

Empirical characterization of random forest variable importance measures

Microarray studies yield data sets consisting of a large number of candidate predictors (genes) on a small number of observations (samples). When interest lies in predicting phenotypic class using gene expression data, often the goals are both to produce an accurate classifier and to uncover the predictive structure of the problem. Most machine learning methods, such as k-nearest neighbors, support vector machines, and neural networks, are useful for classification. However, these methods provide no insight regarding the covariates that best contribute to the predictive structure. Other methods, such as linear discriminant analysis, require the predictor space be substantially reduced prior to deriving the classifier. A recently developed method, random forests (RF), does not require reduction of the predictor space prior to classification. Additionally, RF yield variable importance measures for each candidate predictor. This study examined the effectiveness of RF variable importance measures in identifying the true predictor among a large number of candidate predictors. An extensive simulation study was conducted using 20 levels of correlation among the predictor variables and 7 levels of association between the true predictor and the dichotomous response. We conclude that the RF methodology is attractive for use in classification problems when the goals of the study are to produce an accurate classifier and to provide insight regarding the discriminative ability of individual predictor variables. Such goals are common among microarray studies, and therefore application of the RF methodology for the purpose of obtaining variable importance measures is demonstrated on a microarray data set.

Partial breast brachytherapy after lumpectomy: Low-dose-rate and high-dose-rate experience

PURPOSE The use of partial breast brachytherapy (PBB) after lumpectomy for selected patients with early-stage breast cancer reduces the adjuvant radiotherapy treatment time to <1 week. Despite the advantages of accelerated treatment, maintaining an acceptable cosmetic outcome is important. In a cohort of patients who received low-dose-rate (LDR) or high-dose-rate (HDR) PBB after lumpectomy, the clinical characteristics and treatment parameters were analyzed to identify predictors for an unfavorable cosmetic outcome. METHODS AND MATERIALS Early-stage breast cancer patients with clear resection margins and 0-3 positive lymph nodes were eligible for PBB. Uniform guidelines for target definition and brachytherapy catheter placement were applied. The HDR PBB dose was 34 Gy in 10 fractions within 5 days, and the LDR dose was 45 Gy given at a rate of 50 cGy/h. The end points included incidence of radiation recall reaction, telangiectasias, and cosmetic-altering fibrosis. RESULTS Between 1995 and 2000, 44 patients with early-stage breast cancer received PBB without adjuvant external beam radiotherapy after lumpectomy (31 HDR PBB, 13 LDR PBB). After a median follow-up of 42 months (range 18-86), all patients remained locally controlled. The overall rate of good/excellent cosmetic outcome was 79.6% overall and 90% with HDR PBB. Radiation recall reactions occurred in 43% of patients (6 of 14) who received adriamycin. LDR PBB and adriamycin were significant predictors for late unfavorable cosmetic changes in univariate analysis (p = 0.003 and p = 0.005, respectively). CONCLUSION Although a high rate of local control and good/excellent cosmetic outcome is provided with HDR PBB, the risk of unfavorable cosmetic changes when treated with both LDR PBB and adriamycin is noteworthy. This suggests that HDR PBB is preferred in patients for whom adriamycin is indicated.

BAALC, the human member of a novel mammalian neuroectoderm gene lineage, is implicated in hematopoiesis and acute leukemia

The molecular basis of human leukemia is heterogeneous. Cytogenetic findings are increasingly associated with molecular abnormalities, some of which are being understood at the functional level. Specific therapies can be developed based on such knowledge. To search for new genes in the acute leukemias, we performed a representational difference analysis. We describe a human gene in chromosome 8q22.3, BAALC (brain and acute leukemia, cytoplasmic), that is highly conserved among mammals but evidently absent from lower organisms. We characterized BAALC on the genomic level and investigated its expression pattern in human and mouse, as well as its complex splicing behavior. In vitro studies of the protein showing its subcellular localization suggest a function in the cytoskeleton network. Two isoforms are specifically expressed in neuroectoderm-derived tissues, but not in tumors or cancer cell lines of nonneural tissue origin. We show that blasts from a subset of patients with acute leukemia greatly overexpress eight different BAALC transcripts, resulting in five protein isoforms. Among patients with acute myeloid leukemia, those overexpressing BAALC show distinctly poor prognosis, pointing to a key role of the BAALC products in leukemia. Our data suggest that BAALC is a gene implicated in both neuroectodermal and hematopoietic cell functions.

11q23 balanced chromosome aberrations in treatment-related myelodysplastic syndromes and acute leukemia: Report from an International Workshop†

Among 511 patients with therapy‐related myelodysplastic syndrome or acute leukemia (t‐MDS/t‐AL) and balanced chromosome aberrations, 162 (32%) had translocations involving 11q23. The recurring translocation partners were 9p22 (48%), 19p13.3 (11%), 19p13.1 (10%), 4q21 (9%), 6q27 (6%), 1p32 (2%), 16p13.1 (2%), 10p13 (1%), and 17q25 (1%); in 9%, the translocations were seen only once. The remaining 349 patients were divided into five subgroups based on the balanced aberration: 21q22, inv(16), t(15;17), Rare, and Unique aberrations. Patients in the 11q23 subgroup had a sole cytogenetic abnormality more often than those in the 21q22, inv(16), Rare, and Unique subgroups, and a complex karyotype or −5/del(5q) and/or −7/del(7q) less often than patients in the 21q22, Rare, and Unique subgroups. Clinically, 11q23 patients had acute lymphoblastic leukemia (ALL) more often as their primary disease and a shorter latency from start of treatment for the primary disease to their t‐MDS/t‐AL diagnosis, except when compared with the inv(16) subgroup. The 11q23 subgroup demonstrated a younger age at t‐MDS/t‐AL diagnosis, but this finding was not significant when patients with AL as their primary diagnosis were excluded. Survival from the time of diagnosis of t‐MDS/t‐AL was significantly shorter for the 11q23 subgroup compared with that of the 21q22, inv(16), and t(15;17) subgroups (median 8 vs. 14, 28, and 29 months, respectively). Inferior survival occurred even though 11q23 patients were younger and more often received blood or marrow transplantation (BMT). Even among patients receiving BMT, 11q23 patients had a shorter median survival (9 vs. 12–31 months for the other subgroups). However, among 11q23 patients, those receiving BMT survived longer, with 1‐ and 5‐year survivals of 43% and 18% compared with 23% and 7% for patients not transplanted. With regard to prior therapy, 11q23 patients, compared with other patients, received radiotherapy less often as their sole therapy and chemotherapy more often. They had received VP16, methotrexate, 6MP/6TG, L‐asparaginase, daunorubicin, cytarabine, and VM26 more often, likely attributed to the high frequency of AL as their primary disease. More patients in the 11q23 subgroup had received doxorubicin, except in comparison with the 21q22 subgroup; more vincristine, except in comparison with the Rare and Unique subgroups; and more prednisone, except in comparison with the Unique subgroup. Patients in the 11q23 subgroup more often received alkylating agents (AAs) (86% vs. 59–82% for the other subgroups), and topoisomerase II inhibitors (TIs) (84% vs. 49–75%), and they more often reported exposure to AAs plus TIs without radiotherapy (33% vs. 12–21%), except in comparison with the 21q22 subgroup (36%). We performed a multivariate analysis to determine whether the adverse survival of 11q23 patients compared to other Workshop patients was explained by factors other than the presence of the 11q23 abnormality. Covariates in the final model were the five cytogenetic subgroup indicators, where the 11q23 subgroup was the referent (P < 0.0001); age at t‐MDS/t‐AL (P = 0.0036); previous exposure to lomustine (P < 0.0001) and mitoxantrone (P = 0.0225); BMT for t‐MDS/t‐AL (P = 0.0006); and karyotype complexity (P = 0.0114). The risk of death for 11q23 patients relative to patients in the 21q22, inv(16), t(15;17), and Unique subgroups was significant, even after adjustment for other risk factors (relative risks 2.3, 3.6, 3.1, and 1.5, respectively; P < 0.0001 for the first three comparisons and P = 0.0125 for the last). When a multivariable model was constructed, excluding patients with AL or MDS as their primary diagnosis, the relative risk of death for 11q23 patients was significantly higher than that of all five other cytogenetic subgroups. We conclude that among t‐MDS/t‐AL patients with balanced aberrations, 11q23 translocations are an independent adverse risk factor. Although BMT is the current therapy of choice, new treatment is required.

Depsipeptide (FR 901228) promotes histone acetylation, gene transcription, apoptosis and its activity is enhanced by DNA methyltransferase inhibitors in AML1/ETO-positive leukemic cells.

In t(8;21) acute myeloid leukemia (AML), the AML1/ETO fusion protein promotes leukemogenesis by recruiting histone deacetylase (HDAC) and silencing AML1target genes important for hematopoietic differentiation. We hypothesized that depsipeptide (FR901228), a novel HDAC inhibitor evaluated in ongoing clinical trials, restores gene transcription and cell differentiation in AML1/ETO-positive cells. A dose-dependent increase in H3 and H4 histone acetylation was noted in depsipeptide-treated AML1/ETO-positive Kasumi-1 cells and blasts from a patient with t(8;21) AML. Consistent with this biological effect, we also showed a dose-dependent increase in cytotoxicity, expression of IL-3, here used as read-out for silenced AML1-target genes, upregulation of CD11b with other morphologic changes suggestive of partial cell differentiation in Kasumi-1 cells. Some of these biologic effects were also attained in other myeloid leukemia cell lines, suggesting that depsipeptide has differentiation and cytotoxic activity in AML cells, regardless of the underlying genomic abnormality. Notably, the activity of depsipeptide was enhanced by 5-aza-2′-deoxycytidine, a DNA methyltransferase inhibitor (DNMT). These two agents in combination resulted in enhanced histone acetylation, IL-3 expression, and cytotoxicity, suggesting HDAC and DNMT activities as a potential dual target in future therapeutic strategies for AML1/ETO and other molecular subgroups of AML.

Goodness-of-fit tests for logistic regression models when data are collected using a complex sampling design

Logistic regression models are frequently used in epidemiological studies for estimating associations that demographic, behavioral, and risk factor variables have on a dichotomous outcome, such as disease being present versus absent. After the coefficients in a logistic regression model have been estimated, goodness-of-fit of the resulting model should be examined, particularly if the purpose of the model is to estimate probabilities of event occurrences. While various goodness-of-fit tests have been proposed, the properties of these tests have been studied under the assumption that observations selected were independent and identically distributed. Increasingly, epidemiologists are using large-scale sample survey data when fitting logistic regression models, such as the National Health Interview Survey or the National Health and Nutrition Examination Survey. Unfortunately, for such situations no goodness-of-fit testing procedures have been developed or implemented in available software. To address this problem, goodness-of-fit tests for logistic regression models when data are collected using complex sampling designs are proposed. Properties of the proposed tests were examined using extensive simulation studies and results were compared to traditional goodness-of-fit tests. A Stata ado function svylogitgof for estimating the F-adjusted mean residual test after svylogit fit is available at the authors website http://www.people.vcu.edu/~kjarcher/Research/Data.htm.

MicroRNA profiles in allograft tissues and paired urines associate with chronic allograft dysfunction with IF/TA

Despite the advances in immunosuppression, renal allograft attrition over time remains unabated due to chronic allograft dysfunction (CAD) with interstitial fibrosis (IF) and tubular atrophy (TA). We aimed to evaluate microRNA (miRNA) signatures in CAD with IF/TA and appraise correlation with paired urine samples and potential utility in prospective evaluation of graft function. MiRNA signatures were established between CAD with IF/TA versus normal allografts by microarray. Validation of the microarray results and prospective evaluation of urine samples was performed using real‐time quantitative‐PCR (RT‐qPCR). Fifty‐six miRNAs were identified in samples with CAD‐IF/TA. Five miRNAs were selected for further validation based on array fold change, p‐value and in silico predicted mRNA targets. We confirmed the differential expression of these five miRNAs by RT‐qPCR using an independent set of samples. Differential expression was detected for miR‐142‐3p, miR‐204, miR‐107 and miR‐211 (p < 0.001) and miR‐32 (p < 0.05). Furthermore, differential expression of miR‐142‐3p (p < 0.01), miR‐204 (p < 0.01) and miR‐211 (p < 0.05) was also observed between patient groups in urine samples. A characteristic miRNA signature for IF/TA that correlates with paired urine samples was identified. These results support the potential use of miRNAs as noninvasive markers of IF/TA and for monitoring graft function.

Genes involved in viral carcinogenesis and tumor initiation in hepatitis C virus-induced hepatocellular carcinoma.

The role of chronic hepatitis C virus (HCV) in the pathogenesis of HCV-associated hepatocellular carcinoma (HCC) remains controversial. To understand the transition from benign to malignant, we studied the gene expression patterns in liver tissues at different stages, including normal, cirrhosis, and different HCC stages. We studied 108 liver tissue samples obtained from 88 distinct patients (41 HCV-cirrhotic tissues, 17 HCV-cirrhotic tissues from patients with HCC, and 47 HCV-HCC tissues). Differentially expressed genes (DEG) were studied by use of high-density oligonucleotide arrays. Among probe sets identified as differentially expressed via the F test, all pairwise comparisons were performed. Cirrhotic tissues with and without concomitant HCC were further evaluated, and a classifier was used to predict whether the tissue type was associated with HCC. Differential expression profiles were analyzed using Interaction Networks and Functional Analysis. Characteristic gene signatures were identified when normal tissue was compared with cirrhosis, cirrhosis with early HCC, and normal with HCC. Pathway analysis classified the cellular and biological functions of the DEG as related to cellular growth and proliferation, cell death and inflammatory disease in cirrhosis; cell death, cell cycle, DNA replication, and immune response in early HCCs; and cell death, cell growth and proliferation, cell cycle, and DNA repair in advanced HCCs. Characteristic gene signatures were identified at different stages of HCV-HCC progression. A set of genes were identified to predict whether the cirrhotic tissue was associated with HCC.

Comparison of Cytogenetic and Molecular Genetic Detection of t(8;21) and inv(16) in a Prospective Series of Adults With De Novo Acute Myeloid Leukemia: A Cancer and Leukemia Group B Study

PURPOSE To prospectively compare cytogenetics and reverse transcriptase-polymerase chain reaction (RT-PCR) for detection of t(8;21)(q22;q22) and inv(16)(p13q22)/t(16;16)(p13;q22), aberrations characteristic of core-binding factor (CBF) acute myeloid leukemia (AML), in 284 adults newly diagnosed with primary AML. PATIENTS AND METHODS Cytogenetic analyses were performed at local laboratories, with results reviewed centrally. RT-PCR for AML1/ETO and CBFbeta/MYH11 was performed centrally. RESULTS CBF AML was ultimately identified in 48 patients: 21 had t(8;21) or its variant and AML1/ETO, and 27 had inv(16)/t(16;16), CBFbeta/MYH11, or both. Initial cytogenetic and RT-PCR analyses correctly classified 95.7% and 96.1% of patients, respectively (P =.83). Initial cytogenetic results were considered to be false-negative in three AML1/ETO-positive patients with unique variants of t(8;21), and in three CBFbeta/MYH11-positive patients with, respectively, an isolated +22; del(16)(q22),+22; and a normal karyotype. The latter three patients were later confirmed to have inv(16)/t(16;16) cytogenetically. Only one of 124 patients reported initially as cytogenetically normal was ultimately RT-PCR-positive. There was no false-positive cytogenetic result. Initial RT-PCR was falsely negative in two patients with inv(16) and falsely positive for AML1/ETO in two and for CBFbeta/MYH11 in another two patients. Two patients with del(16)(q22) were found to be CBFbeta/MYH11-negative. M4Eo marrow morphology was a good predictor of the presence of inv(16)/t(16;16). CONCLUSION Patients with t(8;21) or inv(16) can be successfully identified in prospective multi-institutional clinical trials. Both cytogenetics and RT-PCR detect most such patients, although each method has limitations. RT-PCR is required when the cytogenetic study fails; it is also required to determine whether patients with suspected variants of t(8;21), del(16)(q22), or +22 represent CBF AML. RT-PCR should not replace cytogenetics and should not be used as the only diagnostic test for detection of CBF AML because of the possibility of obtaining false-positive or false-negative results.

Abnormal Cytogenetics at Date of Morphologic Complete Remission Predicts Short Overall and Disease-Free Survival, and Higher Relapse Rate in Adult Acute Myeloid Leukemia: Results From Cancer and Leukemia Group B Study 8461

PURPOSE As most patients with acute myeloid leukemia (AML) with morphologic complete remission (CR) ultimately relapse, better predictors for outcome are needed. Recently, Cheson et al suggested using cytogenetic remission (CRc) as part of the criteria for CR. To our knowledge, ours is the first relatively large study evaluating the usefulness of CRc attained immediately following induction chemotherapy. PATIENTS AND METHODS We included AML patients treated on Cancer and Leukemia Group B front-line studies with cytogenetic samples obtained at diagnosis and at the first day of documented CR following induction. Patients with abnormal cytogenetics at diagnosis, and normal cytogenetics at CR (NCR; n = 103) were compared with those with abnormal cytogenetics both at diagnosis and at CR (ACR; n = 15) for overall survival (OS), disease-free survival (DFS), and cumulative incidence of relapse (CIR). Cox proportional hazards models determined the prognostic significance of cytogenetics at CR, adjusting for other covariates. RESULTS Clinical features were similar for both groups, with the exception of favorable cytogenetics [t(8;21), inv(16)/t(16;16), t(15;17)] at diagnosis, which was more frequent (P =.03) in the NCR group. Median follow-up was 3.1 years (range, 1.0 to 11.4 years). ACR patients had significantly shorter OS (P =.006) and DFS (P =.0001), and higher CIR (P =.0001). In multivariable models, the NCR and ACR groups were predictors for OS (P =.03), DFS (P =.02), and CIR (P =.05). The relative risk of relapse or death was 2.1 times higher for ACR patients than for NCR patients (95% CI, 1.1 to 3.9). CONCLUSION Our data suggest that converting to normal karyotype at the time of first CR is an important prognostic indicator and support the use of CRc as a criterion of CR in AML.