Systems and methods for performing temporal logic queries

Details Systems and methods for performing temporal logic queries Systems and methods for performing temporal logic queries

2001-05-31

2004-05-04

Robinson, Greta (Department: 2177)

Data processing: database and file management or data structures

Database design

Data structure types

C707S793000, C707S793000

Reexamination Certificate

active

06732093

ABSTRACT:

FIELD OF THE INVENTION
The invention relates to computer data retrieval systems, and more particularly to methods for performing temporal logic queries using statistical reliability criteria.
BACKGROUND
In analyzing data about certain real-world problems, it is useful to view the data as temporal data. Temporal data typically describe how an observed quantity changes over time or when certain conditions occur. Temporal data are useful for answering questions such as “how often,” “how long until,” or “is this getting better with time.”
Examples of temporal data and possible questions that could be asserted against that temporal data include the following:
Dates of heart attacks (does treatment X reduce the risk of fatal heart attacks?)
Monthly blood pressure readings (does treatment X reduce high blood pressure?)
Periods of hospitalization (how long are patients with condition X staying in the hospital?)
Arrest records (does community policing have a favorable effect? for how long?)
Point-of-sale information for a retail store (which ad campaigns improved sales?)
Temporal data may be organized into “chronologies,” which are sequences of related temporal data. For “period chronologies” each datum has a start- and end-time. For “event chronologies,” each datum has an “event” time. Each datum usually has associated information. The sequence of hospital stays for a patient would be a period chronology, and its associated data might include information regarding diagnosis, facility, and admitting physician. The dates of heart attacks for a patient would be an event chronology, and its associated data might include a severity score.
Temporal data may also contain “eternal” data. These are temporal elements that are time-invariant, per-patient data. Eternal data is true for all observed time. For example, gender, county of birth, date of birth, date of death, index date, censoring date, etc. are stored as eternal facts. These eternal facts have proven useful in temporal logic manipulations, and also provide a way of packaging some of the “critical” data, such as an index or right-censor date, as well as computationally useful data, such as date-of-birth, which is used to convert date-of-event into age-at-event. Such temporal objects thus include: period chronologies, event chronologies and eternal facts. These are all segregated by patient. Additional patient-independent objects may also be provided.
Temporal logic is a useful way of manipulating temporal data. Two periods can have a relationship to each other. For example, two periods can have a containment relationship, where period A completely contains period B, or an overlap relationship, where period A and period B partially overlap each other. Temporal logic extends these notions by providing operations that combine temporal data. For example, if it is desirable to know which patients had heart attacks while being treated for diabetes, this can involve an operation that combines temporal information about heart attacks with temporal information about periods of treatment for diagnosis.
Table 1, below, shows some representative operations and their application.
TABLE 1
Convert event to time
Find the period “30 days before the first heart
period
attack”. This uses the event “first heart attack” as
the end of a period of 30 days.
Invert time period
Find periods when a patient was not on
antibiotics.
Intersect time periods
Find periods when a patient was on beta-blockers
and antibiotics.
Operations can be cascaded. For example, “patients who were not on beta blockers in the 30 days before a heart attack” can be identified by:
converting the heart-attack event to a 30-day period (“convert event to time period”),
finding the periods when the patient was not on beta blockers (“invert time period”), and
Intersecting the two resulting chronologies (“intersect time periods”).
Temporal data introduces a rich variety of complexities into analysis. Causality is often associated with temporal order. Many scientific studies are based on the paradigm “intervene, then observe what happens afterward.” Similarly, with temporal data assertions can often be made about duration: “the patient was pain-free for 4.5 hours.”
Implicit in analyzing temporal data are the concepts of truncation and censoring: data is often gathered from only a limited period of time. Knowledge of what happened before observation begins is limited, and is probably completely unknown after the period in which data is gathered. This phenomenon is called “truncation” of data, referring to the period of observation for the aggregate of units (e.g. subjects, participants, or patients). “Censoring” refers to starts and stops of observation at the level of the unit. Thus, individual units may have varying lengths of observation, which can profoundly alter techniques appropriate for statistically valid uses of temporal data.
Consider a study comparing two methods for preventing heart attacks in high-risk patients. The patients are treated and then monitored for a finite period of time, such as four months. The data would be left-truncated before treatment and right-truncated after the four months. At the end of the study, it is desired to compare the average time-to-heart-attack for the two treatment groups. During the four months, some patients will have reported a heart attack. Assuming they have had continuous observation since treatment began, their contribution to the average time-to-heart-attack for their treatment group is obvious. However, special considerations may have to be given to the patients who may have not been under observation for the entire four months duration. Perhaps some patients died, dropped out of the study or migrated from the area. Their survival time during the four months must be measured, but it's often not clear how their observation time should be incorporated into the comparison of average time-to-heart-attack. Using four months for each of them is plausible but may underestimate the true average time-to-heart-attack, just as assuming that they will never have an attack overestimates the time-to-heart-attack. This lack of data throughout the aggregate observational duration is termed “censoring.” Specifically, right-censoring occurs when patients are observed from the left-truncation date, but are not monitored through the entire period up to the right-truncation date. When individual units are observed until the aggregate right-truncation date (such as the official end of the study), their right-censoring and right-truncation dates are identical, and in practicality synonymous.
A similar situation can occur when the time after the beginning of aggregate observation is relevant to an analysis, but the observed or collected information about the past is subject to a “begin date” cutoff. This is called “left censoring,” and may occur, for example, when a patient is a late entry into an ongoing study.
The heart attack study example also introduces the notion of an index date, which is the date that a subject formally entered the study. The time after the index date is often called the “longitudinal follow-up period;” as noted, this time period can be limited by right-censoring, and is ultimately limited by right-truncation for the aggregate.
Information pertaining to the period before the index date is often called the “baseline period.” The baseline period usually extends only a limited time into the past, for various reasons such as: lack of data, limitations of recall, or because the earlier data is not informative.
Critical data, such as the date data discussed above, depend on factors such as the design of the data or study or the data subject (e.g. patient). The left- and right-truncation dates, for example, depend on the data or study design. Each individual in a study cohort may have different index, right-, and left-censor dates, yielding differing durations of longitudinal follow-up.
Statisticians have developed methods for the analysis of temporal data, which is often referred to as “analysis of lon

Affiliated with

Also associated with

Rating

Systems and methods for performing temporal logic queries does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Systems and methods for performing temporal logic queries, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Systems and methods for performing temporal logic queries will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3194826