Working from Claims Data - Text Mining Techniques for Healthcare Provider Quality Determination - page 342

would not be recoverable. However, inpatient medications are usually not identified separately in the
billing record, so such information would still remain incomplete.
Moreover, although hospital providers tend to use ICD9 codes, physicians are more likely to use the
CPT coding system. Therefore, in addition to defining episodes, it is necessary to be able to use multiple
coding systems. In this chapter, we will discuss the means of working with such data to determine patient
severity. Because of the nature of text mining, it is very well equipped to use multiple coding systems
since it mostly relies upon the linkage between codes instead of relying on a specific list of codes. It is
the linkage between codes that is used in text mining to define patient clusters.
Background
Claims data are used to identify general trends concerning patient treatment. (Smith-Bindman, Quale,
Chu, Rosenberg, & Kerlikowske, 2006; Wilkinson, Askew, & Dixon, 2006) Claims data are also used to
make cross-references to different treatments. In Ceratti, Roger France, and Beguin, claims information
was compared to the hospital clinical database to determine if treatment and diagnosis were related, or
whether treatment was absent given a diagnosis. (Ceratti, France, & Beguin, 2008) Such observational
studies relating billing data to clinical outcomes are fairly common.(Kaushal, Bates, Franz, Soukup, &
Rothschild, 2007) Billing charges for prophylactic medication can also be compared to average costs
of disease treatment.(Collinet-Adler et al., 2007)
Another study cross-referenced billing data from general practice to that of hospital emergency de-
partments to see if patient visits for infection in the general practice could predict near term increases
in emergency room utilization for similar infections.(Sloane et al., 2006) Another use of claims data
has been to determine whether appropriate testing is conducted for patients with chronic illnesses. For
example, claims are used to determine if patients with diabetes are administered regular A1C tests or to
determine if patients with heart conditions are prescribed an ace inhibitor. (Philipneri et al., 2008)
Since multiple providers treat the same patients, the development of a patient severity index will not
be as useful to rank the quality of providers since it will become difficult to separate the contribution of
each provider. Instead, it can be used to find patterns of treatment and to find those treatment patterns
that lead to the best patient outcomes. It can be used to find the relationship between different treatments
and different outcomes for the same patient conditions.
First, the data have to be examined to define separate episodes, and to investigate episodes in relation-
ship to patient outcomes. Solutions under the general category of episode grouper have been developed
specifically to fuse claims data. The methodology is difficult to find since it is mostly proprietary, and
little exists in the research literature.(P. Claus, P. Carpenter, C. Chute, D. Mohr, & P. Gibbons, 1997;
Forthman, Dove, & Wooster, 2000; Rosen & Mayer-Oakes, 1999) A brief summary is given in Forth-
man, Dove and Wooster.(Forthman et al., 2000) The main purpose of these groupers is to identify
homogeneous groups of patients so that cost comparisons and summaries can be made. These “episode
groupers” are used in analysis with little understanding as to how episodes are defined or how patients
are grouped.(Bassin, 1999; Currie et al., 2005; Kerr, McGlynn, Vorst, & Wickstrom, 2000; Thomas,
2005; Wan, Crown, Berndt, Finkelstein, & Ling, 2002) However, it is known that the groupers do not
take into consideration the severity of an individual patient's condition.(Thomas, 2005)
One method of grouping is to examine medications of a similar nature, and to define the end of an
episode if there is at least one day between prescriptions.(Bonetto, Nose, & Barbui, 2006) The Medicare