Physician Practice and Disease-Specific Applications

Abstract

Physician practice applications to date have been limited due to the complicated nature of accounting for the performance of physicians with available data. Physician practice applications are often referred to as clinical applications or primary care physician models in the literature. Although physician practice on specific diseases is often the main focus of the applications in this area (Ozcan 1998; Ozcan et al. 2000), more generic models of physician production were also modeled (Chilingerian and Sherman 1997b).

Appendix K: CPT-Based Claim Processing and Data Development (Source: Ozcan 1998)

1.1 K.1 Procedures for Development of an Episode

The development for CPT-based severity classification requires structuring multiple visits to a particular provider within the time frame (i.e., year). For example, a patient could have two visits to one provider, and 6 months later he/she can go to another provider. Thus, all multiple claims to a specific provider should be aggregated to a single-level claim file for the primary care provider (PCP).

In the subsequent stage of data structuring, claims would be sorted based on the recipient’s age (in terms of calendar days) at the time of service and the claim patterns would be examined. In the chronology of events based on aggregated provider claims, different patterns would be identified to develop decision rules for the identification of episodes, hence the inclusion/exclusion of various claims to the final database. Decision rules associated in this stage include:

• In the chronology of the events, if non-physician claims were preceded by a physician claim, and they were more than the usual episode time apart (2 months for otitis media), they can be ruled out as the end of an earlier episode which the provided time window did not have sufficient information to build as part of the current episode.

• If no physician claim is encountered, then the whole claim stream for the particular recipient can be deleted.

These decision rules enabled physician claims to serve as the trigger for the start of an episode. On the next level, the decision rules assessed whether the encounter with the physician was with a PCP or a specialist. One can also observe that there could be instances when specialists are acting as PCPs.

Once the PCP is identified in the claim stream, all claims should be followed in the chronology of the claims for the recipient that were attributed to that PCP. These should include referrals to specialists, ER, inpatient hospitalization, pharmacy, and lab claims. The same patient, however, could change his/her PCP in time and go to another PCP in a different time window within the evaluation period. Claims following such instances can be attributed to those PCPs who were taking care of the new patient, hence the start of a new episode. If a claim was filed by a specialist following a claim by an internist, pediatrician, or family/general practitioner, this particular claim can be attributed to a PCP as part of specialist use in the treatment of care.

Since the unit of analysis for the evaluation is the PCP, the final aggregation of the data should be conducted by identifying a number of recipients for each physician who acted as the PCP for the patient’s episode. This way, for the relevant disease, patient panels for each PCP during an evaluation period can be identified (Ozcan 1998).

1.2 K.2 CPT Code Creep

The prospective payment system (PPS) has always depended on the accurate reporting of clinical diagnoses and procedures. If errors are present in the reporting process, over-reimbursement or under-reimbursement of services can occur. Since the implementation of the PPS, there have been increases in the average case mix index. Because each percent increase in the case mix index corresponds to large growth in revenue for providers (Carter et al. 1990), this increase has been closely examined.

Payers of health care services are concerned that a majority of this change is due to upcoding. Upcoding, or code creep, is when a provider bills for services that are more extensive or intensive than the ones really performed; services are coded as higher weighted diagnoses, tests, or procedures, when there is no change in the actual resources needed or used. Payers believe that much of the case mix increase has occurred because the PPS gives providers an incentive to code more completely, and in cases of ambiguity, to assign the most highly weighted or complex diagnosis or procedure as principal. By contrast, providers have argued that most of the change in the case mix index is correct and reflects a mix of more complex cases. They believe the increase has transpired because with the implementation of managed care, the less complex, lower weight cases have been moved outside the traditional medical setting.

Recent evidence tends to illustrate that most of the rise in the case mix index is an accurate reflection of increasing complexity (Carter et al. 1990). Studies have found that, on average, only one-third of claims have coding errors (Bailey 1990; Hsia et al. 1992; Siwolop 1989; Shwartz et al. 1996) and that this number can fall below 10 % when dealing with some CPT codes (Javitt et al. 1993). Providers may have under-coded prior to PPS because it made little difference in their payment. The implementation of PPS may have incented doctors to be more accurate about diagnosing and classifying procedures in order to get the proper reimbursement. Nevertheless, in order to overcome the possibility of CPT code creep, we devised an adjustment algorithm so that the severity of the patients was close to actual occurrences.

1.3 K.3 Adjustment Algorithm

The number of PCPs identified needs to be clustered for post-hoc evaluation based on the severity weight class of the patient panels they have seen during the year. The clustering can be done using an index of severity measures that incorporates a weighted volume of patients from each severity class. More specifically, it is assumed that the PCP’s workload for the patients in the second tier of severity would be three times as much, relative to the first peer. Similarly, the PCP’s workload for patients in the third tier would be three times that of the second tier’s severity, or nine times those of the first tier’s severity. Using the following weighing formula, each PCP’s volume/severity workload can be indexed (I_i), and the cluster weight (C_j) for each PCP can be calculated as follows:

$$ {I}_i=\frac{w_i}{{\displaystyle \sum_{i=1}^n{w}_i/n}}\kern2em i=\left(1,\dots, 3\right) $$

$$ {C}_j={\displaystyle \sum_{j=1}^m{P}_{ij}*{I}_i}\kern2em \left(i=1,..,3;\kern0.5em j=1,\dots, m\right) $$

where P_ij represents the number patients in the ith class of severity for the jth PCP.

As an example, the index values (I_i) calculated in Ozcan (1998) using 160 PCPs were 0.23, 0.69, and 2.08 for respective severity categories. Cluster weight distribution (C_j) ranged from 0.2016 to 0.7479. Results of the adjustment algorithm showed 6 PCPs were classified in the high severity/volume cluster, 22 PCPs in the medium, and the remaining 132 PCPs were designated to the low severity/volume cluster. This representation fit reasonably well to expectations.