Loading...

Introduction

The PCATS online App is designed to offer an online graphic user interface (GUI) that implements Bayesian causal inference methods for various types of treatment encountered in comparative effectiveness research. Here, the term treatment is used in a broader sense, referring to the actions or exposures that are expected to generate different outcomes. Some examples of a treatment include medication therapy to treat an illness, a diet plan to manage weight gain, or an educational program to improve certain knowledge/skills. The simplest type of treatment is a binary dichotomize treatment, where only two choices of the treatment are considered, such as a new intervention vs. a conventional control. However, most of the time, the treatment cannot be characterized in this simple way, rather, treatment is much more intricate. A treatment could have more than one categories. For example, a diet intervention could be vegetarian, Paleo, low carb or Atkins diet. It could be combination of multiple treatments, such as a weight management program that includes both diet and exercise. Additionally, treatment could be continuous, such as number of fruit serving or counting of steps taken each day. Furthermore, a treatment could change over time. For example, a cancer patient may undergo a surgery first before going through chemotherapy. The time varying treatment can be adaptive or non-adaptive. The adaptive treatment is not determined ahead of time, but rather adapts to what happens as treatment progresses, thus treatment is constantly changing over time. The non-adaptive treatment is determined ahead of time, and adjustments over time are pre-planned regardless of what happens after the treatment. Most of the commonly used causal inference methods consider the simple non-adaptive binary type of treatment. The PCATS App implements Bayesian's nonparametric causal inference approaches considering both adaptive and non-adaptive types of treatment, can be applied to both experimental and nonexperimental observational data.

PCATS stands for patient centered adaptive treatment strategy. The PCATS project was motivated by comparative effectiveness researches aiming to find better treatment strategies for patients with chronic illnesses. Routine clinical approaches seeking to treat patients with chronic or prolonged disease conditions are adaptive, where the patients often go through many courses or stages of treatment. Often, it is unclear as to what treatment plan works the best, and how the treatment should be adjusted over time. Yet, such knowledge could make enormous difference in helping patients. Motivated by the patient centered comparative effectiveness questions, specifically "what is the best 1st line treatment option for a patient like me", "what is a better next treatment option given my past treatment history and disease progression", PCATS study is aimed to understand effectiveness of complexed treatment types utilizing data collected from real clinical encounters by: 1) refine and improve Bayesian causal inference methods; and 2) evaluate effectiveness of different time-varying adaptive treatment strategies for treating children with newly diagnosed juvenile idiopathic arthritis (JIA).

Usually patient report outcomes are obtained via questionnaires and is bounded by the minimum and maximum possible values. Many clinical outcomes measured also subject to the ceiling or floor constrains. For example, laboratory measures are usually limited by the lower and/or upper detection limit. Physician global evaluation of the patient disease activity is bounded between 0 and 1. The bounded feature of an outcome measure requires appropriate consideration in the CER. Failure to do so could bias the study results. Therefore, the PCATS App offers the option for users to specify outcome bounds, and implemented method properly evaluate bounded outcomes.

This online application implements two Bayesian's nonparametric causal inference methods: GPMatch for continuous outcome, and BART for non-continuous outcome, including binary and counting type of outcomes.

The PCATS App offers many advanced features for implementing robust causal inference with:

1) complex treatment, including binary, multilevel, continuous and composite, and time-varying adaptive treatment, as well non-adaptive treatment strategies;

2) post-treatment confounding factors (under construction), such as adherence;

3) heterogeneous causal treatment effect; and

4) bounded outcome.