UCLA Department of Statistics

Projects

Susan Love Foundation

Our consulting team worked with the Dr. Susan Love Research Foundation on two projects. The first was to model the levels of four hormones and residual proteins in the breast ductile lavage fluid of healthy women. Methods used include nested models, data transformation, simulating missing data and intraclass correlation (ICC). The second was to explore chemical levels collected of patients over several hours after exposure to different medications, comparing levels between ductile lavage, blood and nipple aspirative fluid (NAF) with plots. Methods include ggplot, time series and kernel smoothing estimation.

Click here for completed report.

Neonatal Candidiasis Risk Factors

Methodology: Logistic regression
Client affiliation: UCLA Medical Center, Pediatric Infectious Diseases Clinic

Medical researchers are studying the effect of treatment delay and various other factors in neonatal candidiasis. We conducted logistic regression analysis to determine how treatment delay and other various factors affect the likelihood of survival of the subject. We also investigated the relationship between these factors and the outcome when patients whose only culture was from a respiratory source were excluded from the analysis.


Click here for completed report.

CVCB Antimicrobial Lock Therapy Study

Methodology: Pearson’s chi-square test and generalized linear models
Client affiliation: Stamford Hospital, Department of Infectious Diseases and Microbiology

Episodes of central venous catheter-related bacteremia (CVCB) have become common events in patients with long-term indwelling central venous catheter use. Since no standardized treatment of CVCB exists, novel approaches have been used to salvage infected catheters where vascular access is limited. Use of antibiotic lock solutions (e.g., a solution of an antibiotic agent and heparin) to fill catheter lumens between treatments (antibiotic-lock treatment, ALT) has been shown to prevent CVCB and may be useful for treatment of already-infected catheters. In this report we analyze the success of ALT in preventing CVCB.


Click here for completed report.

Type 2 Diabetes Mellitus and Melatonin Levels

Methodology: Power analysis
Client affiliation: UCLA Medical Center, Division of Endocrinology

We provided medical researchers planning a study investigating the melatonin levels of patients with type 2 diabetes mellitus with a power analysis in order to help them determine the sample size required for certain effect size and power combinations.


Click here for completed report.

Assersion of comprehensive approach to dermatology

Methodology: Multivariate regression, Bayesian statistics
Client affiliation: UCLA Medical Center, Division of Dermatology

Study was conducted to see if comprehensive approach to dermatology is more effective than treating the skin condition alone with topicasls. Subjects were divided into the treatment and control group and 10 minute educational intervention was given to the treatment group along with topical corticosteroids and just topical corticosteroids and no intervention to the control group.
Several kinds of measurement was taken before and after the intervention for the treatment group and control group of which 4 are of primary interest; phase angle, basal metabolic rate, intracellular water, and free fat mass. Result showed treatment effect in the expected direction, yet because of the small sample size result showed that treatment was not statistically significant.


Click here for completed report.

Carpinteria Childbearing Survey

Methodology: Exploratory data analysis, data visualization
Client affiliation: First 5 Santa Barbara County Early Care and Education Division

A voluntary survey on women in Carpinteria about their childbirth experience was conducted to help design a program to benefit pregnant women and new moms. This report summarizes the results of this survey.


Click here for completed report.

Thyroid Cancer and Iodine Levels

Methodology: Power analysis
Client affiliation: UCLA Medical Center, Division of Endocrinology

We provided medical researchers planning a study investigating iodine levels of thyroid cancer patients with a power analysis in order to help them determine the sample size required for certain effect size and power combinations.


Click here for completed report.

Surgery preference of orthopedic surgeons based on X-rays

Methodology: Generalized linear mixed effects models
Client affiliation: UCLA Medical Center, Department of Orthopaedic Surgery

For this project we are working with orthopedic surgeons from UCLA Medical Center who conducted an online survey that 700+ orthopedic surgeons responded to. In this survey the surgeons were asked about their training (3 categories), their practice (academic vs. private) and what percentage of their practice involves treatment of pediatric patients (4 categories). Then they were shown a series of 10 X-ray images and asked whether they would choose to operate on the patient based on the X-ray. Multi-level modeling is used to examine if there are significant differences in rate of surgery preference based on training, practice and percentage of treatment of pediatric patients the surgeon’s practice involves.


Click here for completed report.

Quality of Reproductive Health Services to Limited English Proficient Patients

Methodology: Generalized linear mixed effects models
Client affiliation: Bixby Center for Global Reproductive Health

This study explores the association between language discordance and key indicators of reproductive health care quality. We seek to test the hypothesis that patients seen by a bilingual provider, on average, receive higher quality services than patients who receive reproductive health services with the help of an interpreter. In order to do so we have analyzed patient care indicator data using generalized linear mixed models with nested structure.


Click here for completed report.

Executive Compensation

Methodology: Multinomial logistic regression, generalized linear mixed models, linear regression with random intercept
Client affiliation: Boston University School of Law

Analyzing data from fortune 1500 companies over two recent years for the five top executives within each company. The data includes variables both at the individual executive level and at the company level. The outcome of interest is the proportion of compensation in stock and stock options composed of stock options. We will merge data from two financial databases and fit an appropriate multilevel/repeated measures model for this data.


Click here for completed report.

Identification of Optimal Buy and Sell Periods

Methodology: Financial data simulation and forecasting
Client affiliation: Fried Asset Management, Inc.

In this stocks project, we examined a client’s stock portfolio over a decade. We looked at the distribution of closing prices among the holdings and calculated the returns that would be seen with various combinations of hard sell points and trailing stop losses. Moving averages over 50 and 200 days were analyzed in the hopes of identifying optimal buy and sell periods based on these statistics.

Surgical Drainage Meta-Analysis

Methodology: Meta-analysis
Client affiliation: Aesthetic and Plastic Surgery Institute, University of California, Irvine

Surgeons routinely use prophylactic drainage after surgery however this practice is often disliked by patients and the benefits are not always clear. Recently surgeons have been explicitly testing the value of drainage in specific surgical proceedures. The objective of this project is to determine the evidence-based value of prophylactic drainage in surgeries where drainage is above the fascia and below subcutaneous tissue. We are performing a meta-analysis of existing studies involving these kinds of drainage to characterize the drain effectiveness using a random-effects model considering possible outcomes such as hematomas, abscesses, seromas and or infections.


Click here for completed report.

Application of Constrained PCA to Understanding Brands

Methodology: Constrained PCA
Client affiliation: A marketing research firm in San Diego

The objective of this project is to develop an algorithm for Constained PCA that will be used in to learn about positioning as well as weaknesses of brands to be used by a marketing research firm.


As a final product R code was delivered to the client.

Evolution of Channel Islands Population

Methodology: Homogeneity analysis using HOMALS in R
Client affiliation: California State University – Los Angeles, Department of Physical Anthropology

There has been population replacement during the prehistoric times on the Channel Islands. Before the Spanish came, there were two different Indian populations: one in the North and one in the South of the Channel Island, confirmed by cranial measurements and Mitochondrial DNA (MDNA). The Northern population (is closer to Santa Barbara) are the Chumash Indians and the Southern one, the Ituaztecan. The Ituaztecan spread to Central America and pushed out the Chumash. (We do not know around what time this happened.) Using MDNA, the client has 6 reliable skull classifications (from 3 different islands), belonging to either of the two groups. With this information, he would like to be able to classify about 100 other individuals, based on their cranial measurements, taking location into account.


Click here for completed report.

Fuel Loss

Methodology: Exploratory data analysis, non-parametric regression
Client affiliation: United Oil

In this project we investigate transportation fuel loss among retailers and what environmental elements or business practices might influence such fuel loss. The transportation fuels industry has monitored the disappearance of a small percentage of fuel throughout retailers in Santa Barbara, Los Angeles, Orange County, and San Diego since 2001, and we will investigate potential causes of this loss to help mitigate future losses.