Coding Risk Scores

Critical for Fiscal Survival

Executive Summary

By Brad Sweet
IRP Systems, Inc.

Improving coding quality has been one of the traditional goals of Medical Records, especially since the introduction of Diagnostic Related Groups (DRG's) in the Medicare Prospective Payment System (PPS) in 1983. Since coding quality impacts DRG assignment and reimbursement, correct and complete coding is critical for fiscal survival, both today and in the future under proposed health care legislation.

Information Resource Products, Inc., the parent company of IRP Systems, Inc. has been directly involved with coding and reimbursement issues since 1984 and is currently conducting a national study concerning complete coding and its impact on DRG reimbursement. In this study, IRP looked at several million Medicare discharges and developed a computer model which analyzed why hospitals with similar patient mixes are reimbursed differently.

THE QUANTITATIVE results of this computer analysis is a Coding Risk Score which is used to rank cases for coding review. This score takes into consideration the probability that additional code(s) exist for a particular chart, but have not been coded for one reason or another. This score also factors in the increase in DRG weight of the resulting new DRG. In general, a higher Coding Risk Score strongly suggests that a review of the chart is likely to yield coding improvements resulting in increased DRG payments.

Needless to say, the formulas used to calculate the Coding Risk Score are very complicated and detailed. In fact, IRP's DRG/Expert computer model has identified approximately 6000 expert rules for scoring cases for possible coding improvement. However, the following sample illustrates some of the concepts involved in this process.

Every year there are between 400,000 and 500,000 cases of simple pneumonia with 90 percent of these cases having one or more complications/comorbidities which results in DRG 089 Simple Pneumonia & Pleurisy Age > 17 w/CC. DRG 089 has a reimbursement rate which is 62 percent higher than its non-CC counterpart, DRG 090.

IF ONE WERE to select charts of pneumonia patients for a coding review, those charts without a complication should be reviewed first based upon the high probability of a complication/comorbidity. Reviewing a chart for an unreported complication is just one of several review steps that may be performed during a coding quality audit.

Taking into account the principal diagnosis, further refinements can be made because statistically the bacterial pneumonia diagnoses have a significantly higher rate of complications/comorbidities than do the viral variety. In fact, bacterial pneumonia cases typically have two or more complications/comorbidities on the average, based upon the computer model analysis.

The computer model further refines the coding risk score by taking into account the patient's sex, discharge status and relative age, since analysis has shown that these factors may often influence the likelihood of a complication or other significant diagnosis.

As an example, congestive heart failure is reported 30 percent more often for female patients with bacterial pneumonia than male patients. The corollary of this is that male patients are 30 percent more likely to have chronic airway obstruction reported than their female counterparts. These variations may be the result of lifestyle differences, such as diet, smoking or stress. Genetic differences between the sexes may also be a factor.

ONE LAST refinement used is the impact of actual length of stay vs. the arithmetic length of stay for the particular base DRG. The intent is to help focus the review process on those cases where the actual LOS is greater than the DRG LOS, assuming all other conditions are equal.

The following examples will help to put all of this in perspective.

Let's assume that we had four patients, all males of 67 years in age. Two of the patients had viral pneumonia and the other two had unspecified bacterial pneumonia - all of the cases grouped to DRG 090, Simple Pneumonia & Pleurisy Age > 17 w/o CC. The only other difference between the cases is the actual LOS: one from each group was hospitalized for 6 days, the other 9 days.

THE FOLLOWING data illustrates the coding risk score assigned by the computer model for each of the 4 cases:

                                Risk Score 
                            6 Days    9 Days
                              LOS       LOS
    
viral pneumonia             .9360     1.3371
bacterial pneumonia         1.2358    1.7677

Remember that a higher coding risk score indicates that there is more potential that a case has some unreported codes which will yield a higher weight DRG. Therefore, based upon this potential, the sequence in which these four cases should be reviewed is:

                       LOS       Risk Score
 
bacterial pneumonia    9 days      1.7677
viral pneumonia        9 days      1.3371
bacterial pneumonia    6 days      1.2358
viral pneumonia        6 days       .9360

Estimating the dollar impact of the difference between Coding Risk scores is very difficult and subjective. In the above examples, the difference in revenue between DRG 090 and DRG 089 is approximately $1,700. However, experience shows that only a fraction of cases reviewed will be changed; an estimate is that each tenth (0.1) point difference in code risk scores may have a $100 per case impact on Medicare revenues. Therefore, a hospital with 1,000 Medicare discharges annually, which can reduce its average coding risk score by 0.1, could potentially increase Medicare revenues by $100,000.

THE RESEARCH staff at IRP recently completed a project which involved calculating a coding risk score for every inpatient Medicare Discharge for a 12 month period. After running these cases through the computer model, the team summarized the data for a variety of categories including type of ownership, population density, state and bed size range.

Table 1 illustrates the range of coding risk scores by type of ownership and by bed size range within type of ownership. Although there are some differences, the data in this table is fairly consistent across the various groups.

The most notable trend is that the large hospitals, 400-699 beds and especially the 700+ bed facilities, generally have much lower coding risk scores than their smaller counterparts. Since these large facilities perform many surgical procedures, and generally handle more complicated cases, it is plausible that their cases naturally group into more complicated DRGs. Furthermore, these large facilities typically have more sophisticated computer systems and training available to the medical records staff.

ONE INTERESTING variance to that trend is that the larger "non-profit church" and "non-profit other" hospitals don't show as favorable a coding risk score when compared to the "for profit" and "non-profit private" facilities. This can be partially explained by the fact that there are a very large number of "non-profit church" and "non-profit other" facilities in the 400-770+ bed size range.

Overall, the smaller facilities, especially those under 100 beds, have a much higher coding risk score. This may be due to the fact many of these facilities don't have extensive surgical units, or never see the sicker patients who end up at the larger facilities. Often times, the resources available to medical records are limited_ coding may be done manually, their DRG software may be very basic, or their staff may have multiple duties in addition to coding.

These theories are acceptable until you examine the coding risk scores by population density, shown in Table 2. The data in this table paints a much different picture.

IN AREAS WITH a very high (large urban) population density, the 700+ bed facilities actually have a much worse (higher) coding risk score than their counterparts in less densely populated areas. This trend also holds true for the 400-699 bed facilities.

Hospitals in the 200-399 bed range in both the very high and rural areas share the same high coding risk score as the 700+ bed facilities in very high densities.

Hospitals in rural areas, regardless of bedsize, generally have the highest coding risk score. These hospitals also account for about 21 percent of all Medicare claims, while hospitals in the very high areas only account for 15 percent of all Medicare Claims.

Looking at coding risk scores by state presents some startling data. The average coding risk scores are almost three times higher for states at the top end of the scale, as compared to states in the low end of the scale. Table 3 illustrates this phenomenon for two groups of four states each at either end of the scale.

BOTH GROUPS are fairly evenly divided_ the total number of discharges equal about 1.1 million each; both have large states and small states represented; the largest state in each group has traditionally been in the forefront of health care legislation.
Also interesting to note is that there exists wide variances on a hospital by hospital basis within these states.

By using the coding risk score, discharges can be ranked in order by potential coding improvement. A higher score usually indicates that additional codes may exist for a particular chart, but have not been coded for one reason or another.

LET'S LOOK AT two groups consisting of four states each at opposite ends of the coding risk scale. Group 1 (high end) consists of the states of New York, New Jersey, Maine and the territory of Puerto Rico. Group 2 (low end) comprises California, Colorado, Arizona, and the District of Columbia. The combined number of Medicare discharges for these two groups is approximately 2.5 million, which represents nearly 25 percent of all Medicare discharges for one year.

Both groups are similar in that the total number of discharges equals approximately 1.25 million each, very large states and smaller states/territories are represented and the largest state in each group has traditionally been in the forefront of health care legislation.

USING THE DRG/Expert computer model, coding risk scores are assigned to every Medicare discharge in both groups. These scored cases are then grouped by DRG and ranked in descending coding risk score sequence. Selected data from this analysis is presented in Table 4.

The 13 DRG's with the highest coding risk score comprise nearly 700,000 discharges, representing about 25 percent of the 2.5 million discharges for both groups. In addition, the total coding risk score represented by these 13 DRG's is about 50 percent of the total coding risk score for all discharges for both of these groups. The discharges from hospitals in Group 1 consistently have higher coding risk scores than those represented by hospitals in Group 2.

Discharges grouping to DRG 127 - heart failure and shock represent 21 percent of the total discharges for Groups 1 and 2, but account for 27 percent of the total coding risk score for both groups.

THE MOST common coding risk occurs when the principal diagnosis is 428.0 - congestive heart failure or some other form of cardiovascular complication (CVC). Our analysis of Medicare discharges reveals that a good percentage of the time an acute myocardial infarction (AMI) may be present in the chart, but not coded.

When both a cardiovascular complication and an AMI are coded, the case will group to DRG 121 - circulatory disorders with AMI and CVC, assuming the patient is discharged alive. The weight for DRG 121 is approximately 60 percent higher than DRG 127, translating to an average of $2,000 improved Medicare reimbursement. Review of all charts grouping to DRG 127 may prove to be very beneficial.

The second most common coding risk involves cases grouping to DRG 140 - angina pectoris. These cases account for about 12 percent of the total coding risk score, as well as 12 percent of the total Medicare discharges for Groups 1 and 2.

ANOTHER CODING risk occurs when a procedure was performed, but not coded for one reason of another. A common non-operative procedure for similar cases is some form of cardiac catheterization, resulting in DRG 124 with a 100 percent increase in DRG weight as compared to DRG 140. This translates to an average of $3,500 improved Medicare reimbursement.

Additional coding risk occurs when percutaneous transluminal coronary angioplasty (PTCA) is performed, but not coded on the claim form. When this is performed the case will group to DRG 112, which has a 218 percent increase in DRG weight, or $7,500 increase in Medicare reimbursement.

Catching coding deficiencies in cases grouping to DRG 140 has very high payback. However, if your facility is not equipped to perform the specialized procedures which result in either DRG 112 or 124, any review of these cases would be fruitless.

The third most common coding risk involves cases that group to either DRG 088 - chronic obstructive pulmonary disease or DRG 089 - simple pneumonia and pleurisy. Combined these cases account for approximately 16 percent of the of the total coding risk score and 21 percent of the total Medicare discharges for Groups 1 and 2.

A frequent coding risk occurs when an endoscopic lung biopsy was performed, but not included on the claim form. Again, if your facility doesn't typically perform this procedure, then reviewing these cases will probably result in very little pay back.

Cases grouping to DRG 139 - cardiac arrythmia and conduction disorder should be reviewed for a complication/comorbidity (CC), since similar cases generally have a complication reported. The most common CC's reported are 428.0 - congestive heart failure and 496 - chronic obstructive pulmonary disease (COPD). The resulting DRG 138 has an average $2,000 increase in Medicare reimbursement.

CASES GROUPING to DRG 211 - hip and femur procedures except major joint_should be reviewed for a CC as well. The most common CCs are 285.1 acute posthemorrhagic anemia and 496 - COPD. The resulting DRG 210 has an average Medicare reimbursement which is $1,500 greater than DRG 211.

In reviewing the length of stay (LOS) columns in Table 4, it is interesting to note that the average actual LOS days for Group 1 cases is significantly higher when compared to either the average actual LOS days for Group 2 or the HCFA arithmetic LOS days. This trend is consistent for the 13 DRG's with the highest coding risk score.

Since average actual LOS is used by the computer model as one factor in the calculation of coding risk score, this trend may partially explain part of the differences in coding risk scores between Groups 1 and 2.

HCFA ALSO takes into consideration the average LOS when determining whether secondary diagnoses should be added or removed from the list of CCs. Currently this list includes those diagnoses that, when present as a secondary condition, would be considered a substantial complication or comorbidity. In preparing the original list of CC's, HCFA defined a substantial CC as a condition which would increase the LOS by at least one day in at least 75 percent of the patients.

Another factor which impacts the total risk score for a group is the number of cases which group to DRG's having no coding risk. From a DRG standpoint, these cases typically cannot be coded any further to increase Medicare reimbursement.

Table 5 summarizes those DRG's for which no coding risk exists. Group 2 hospitals have on the average 16 percent more cases which have no assigned coding risk, as compared against Group 1. This is another factor that explains the lower coding risk scores for Group 2 hospitals.

In review, we have identified 13 DRG's which account for 50% of the coding risk score for 2 groups of acute care facilities. Several of these DRG's were discussed in detail, showing the coding review opportunities which would have a positive impact on Medicare reimbursement.

For more information, contact IRP's Research Group at (617) 938-6444 ext: 407 or write: Coding Risk Research, IRP Systems, Inc., 5 Alexander Rd, Billerica MA 01821-5032. Requests may also be faxed to (978) 670-9954.