Pediatric Length of Stay Guidelines and Routine Practice: The Case of Milliman and Robertson

Background  Guidelines for inpatient length of stay (LOS) have been developed by Milliman and Robertson (M&R) and are widely applied by health plans. This study was designed to compare LOS for several pediatric conditions with the M&R LOS criteria using recent data and to determine if concordance of actual practice with M&R LOS criteria varied between children and adults.

Design  Administrative data from Pennsylvania hospitals from 1996 through 1998 were used to examine LOS for hospital discharges for 12 selected diagnoses for which M&R published guidelines for children and adults.

Patients  Discharge data for all patients discharged from public and private hospitals in Pennsylvania for which 1 of 12 selected diagnoses were examined.

Main Outcome Measure  Length of stay.

Results  In Pennsylvania hospitals from 1996 through 1998, pediatric LOS was divergent for all conditions examined, although not to the extent found in a previous study examining data from New York State. Of note, median LOS for some conditions was shorter than M&R LOS criteria. The percentage of pediatric hospital discharges that exceeded the M&R LOS criteria ranged from 25% for pneumonia to 84% for meningitis. Adult hospital discharges exceeded M&R LOS criteria to a greater extent than did pediatric discharges for all conditions except for sickle cell crisis and meningitis.

Conclusions  The M&R LOS criteria were divergent from routine practice for both children and adults. Greater divergence of adult discharges illustrates the need to consider comorbid conditions when implementing these guidelines. Thus, patient care may suffer if guidelines are implemented in an uninformed way. These findings emphasize the importance of using the best possible science when producing guidelines such as these.

IN EFFORTS to control costs, many health plans have begun to use clinical practice guidelines to limit use of health services.¹ Although guidelines regarding length of stay (LOS) have been widely disseminated and adopted, there have been few efforts to examine the validity of these guidelines, especially for diverse populations.¹ It is not known for which patients they are most relevant. Previous studies have raised concerns regarding the impact of these guidelines on patients,^2-4 especially since many are not evidence based.

Recently, the LOS guidelines for pediatric conditions published by Milliman and Robertson (M&R),⁵ which are the most widely used by managed care plans throughout the country,⁶ have come under great scrutiny. The M&R guidelines for pediatric LOS were developed by faculty members in the Department of Pediatrics at the University of Texas–Houston Medical School. These guidelines were designed to apply to routine, uncomplicated cases, but there are reports that they are being applied to many cases without regard to complications,⁴ raising concerns about the safety of such guidelines in practice.⁶ In fact, one physician who was credited as a contributing author to the guidelines (without his permission) was quoted as saying, "Kids might die because of these guidelines."⁷ The actual impact of these guidelines on processes of care, however, has yet to be assessed.

In an effort to determine how M&R pediatric LOS guidelines compared with actual LOS among pediatric patients, Sills et al⁴ used data from 1995 from the New York Statewide Planning and Research Cooperative System to examine the percentage of hospital stays and total hospital bed days that exceeded the M&R LOS criteria for 16 diagnoses. They found pediatric LOS to be markedly divergent from the M&R LOS criteria, with a significant proportion of discharges exceeding the criteria. However, it is not clear from their study whether these findings, which were unique to that area (New York State), would persist over time in light of increased managed care penetration and whether they were any less valid for children than adults.

We employed more recent data from a different state (Pennsylvania) to shed further light on the pediatric M&R LOS criteria. We sought to (1) determine if pediatric discharges in Pennsylvania exceeded M&R LOS criteria in similar proportions as those identified by Sills et al,⁴ who found the median LOS to exceed the M&R LOS criteria for 13 of 16 conditions, and (2) compare the proportion of pediatric discharges exceeding M&R LOS criteria to adult discharges exceeding criteria. We hypothesized that the proportion of pediatric discharges that exceeded the M&R LOS criteria would be comparable to that found in the investigation by Sills and colleagues. We also hypothesized that the proportion of discharges that exceeded the adult M&R LOS criteria for those same conditions in children would be even greater because adults are more likely to have medical complications⁸ and thus longer LOS than children.

Data from the Pennsylvania Health Care Cost Containment Council hospital discharge database for the calendar years 1996, 1997, and 1998 were used to determine LOS for children and adults. The database provides comprehensive data for each inpatient hospitalization in a nonmilitary acute care hospital in the state. The study population for this investigation was separated into 2 groups: adults and children. Children were identified as any patient aged 0 to 17 years and adults were identified as any patient aged 18 years or older. Length of stay is defined in the database using a calendar day approach. For example, an admission in the evening and a discharge the next morning would be recorded as a LOS of 2 days, and an admission at 12:01 AM and a discharge at 11:59 PM on that same calendar day would have a LOS of 1 day. Milliman and Robertson LOS criteria are based on LOS calculated using this method.⁵

Patients with a principal diagnosis of appendectomy with complications, appendectomy without complications, asthma, major but noncritical burn, cellulitis, diabetic ketoacidosis, gastroenteritis, bacterial meningitis, osteomyelitis, pneumonia, acute pyelonephritis, and sickle cell crisis were included in the analysis. These conditions were picked to match the set of conditions analyzed by Sills et al,⁴ who chose those conditions to provide a mix of conditions of high and low severity and both medical and surgical conditions. Sills and colleagues also examined appendicitis without abscess, bronchiolitis, croup, and pyloromyotomy, but, because those conditions were extremely rare in adults in our sample, we chose to not examine those 4 conditions in this analysis. Inpatient hospitalizations with the 12 conditions were identified using International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) codes.⁹ The ICD-9-CM codes used are identified in Table 1.

As in the report by Sills and colleagues, we used the functional M&R LOS guidelines as opposed to the goal LOS guidelines. The goal LOS guidelines represent a target for the hospital and are always 1 day shorter than the functional LOS. We used the pediatric LOS guidelines to assess children's hospitalizations and adult M&R LOS guidelines to assess adult hospitalization. There were no differences in the functional LOS guidelines between adults and children for any of the conditions examined.

Because some patients may have required intensive care and had other serious complications during their hospital stay, there is the potential for large outliers with extremely long hospital stays to provide undue weight to the analysis. Therefore, we chose to eliminate outliers from the analysis. For our results to be comparable to the results of Sills and colleagues, we used the same cutoff for eliminating outliers, excluding the 2% longest LOS.

We examined the mean, median, mode, and range of LOS for each of the 12 diagnoses using a computer program (SAS Institute, Cary, NC). We also examined the frequency distribution of LOS for each condition for adults and children in relation to the M&R functional LOS guideline, using histograms with a black bar indicating the M&R LOS guideline. We calculated the percentage of discharges with LOS exceeding the M&R functional LOS guideline as well as the percentage of hospital bed days exceeding the guideline. As the data represent the entire universe of hospital discharges for the 12 conditions in Pennsylvania, no statistical tests are needed or performed. For each set of analyses, we additionally examined the 3 years of data separately to determine if there was an observable time trend and compared LOS for patients with and without a comorbid condition. Any patient with any secondary diagnosis recorded in the discharge record was considered to have a comorbid condition, regardless of the specific secondary diagnosis recorded. In other words, if all of the secondary diagnosis fields on the discharge record were left blank, the patient was considered to have no comorbid conditions, and if anything was recorded in any of the secondary diagnosis fields, the patient was considered to have a comorbid condition.

As given in Table 2, for the 12 conditions examined for children from 1996 through 1998, there were between 164 cases of appendectomy with complications and 22 060 cases of asthma. The modal LOS was at or below the M&R functional LOS guidelines for 10 of the 12 conditions, with modal LOS for major but noncritical burns, pneumonia, and sickle cell crisis falling below the M&R guidelines. Only bacterial meningitis and osteomyelitis had modal LOS that exceeded the M&R guidelines. When the M&R functional LOS guidelines were examined in relation to the median LOS, 7 of the 12 conditions met the M&R guidelines. The median LOS was not below the M&R guidelines for any of the 12 conditions examined. Although only 2 of the 12 conditions had modal LOS that exceeded the M&R LOS criteria in our data, between 29% and 83% of hospital stays exceeded the M&R guidelines. The LOS frequency distributions for the 12 conditions (Figure 1) demonstrate varying proportions of hospitalizations in excess of the M&R LOS guidelines. The M&R functional LOS guideline is imposed on each plot as a vertical bar.

We also examined LOS for each year individually to determine if any time trend was apparent and examined LOS for children with and without recorded comorbid conditions. Although the data did not demonstrate a time trend (ie, the distribution of LOS was essentially the same for each year for all conditions), 63% of children had a comorbid condition. The mean LOS for children with comorbid conditions was 33% longer than the mean LOS for children without comorbid conditions. Longer LOS were seen for children with comorbid conditions for all 12 conditions examined.

To better demonstrate the relationship of M&R guidelines to actual hospital stays, we calculated the total number of hospital stays that exceeded the criteria and the total percentage of hospital bed days that exceeded the M&R LOS criteria for each of the 12 conditions examined (Table 3). Even in the case of pneumonia, for which the LOS comes closest to the guidelines, 29% of all hospital stays by children were in excess of the LOS guidelines. For osteomyelitis, sickle cell crisis, and meningitis, most hospital stays by children exceeded the M&R guidelines, with more than 80% of hospital stays for meningitis exceeding the guidelines. Total bed days exceeded the M&R guidelines by at least 20% for all 12 conditions examined. Furthermore, total bed days exceeded the guidelines by more than 50% for 3 conditions (sickle cell crisis, major but noncritical burn, and meningitis).

For adult hospital stays, the total number of cases ranged from 398 for meningitis to 132 717 for pneumonia (Table 4). The modal LOS exceeded the M&R guidelines for 6 of the 12 conditions, and the median LOS exceeded the guidelines for all but 1 condition (acute pyelonephritis). The extent to which hospital stays exceeded the M&R LOS guidelines for adults is shown in the frequency distribution curves plotted in Figure 1.

The LOS for each year was examined individually to determine if any time trend was apparent. The LOS for adults with and without recorded comorbid conditions was also examined. As was the case with pediatric hospitalizations, the data did not demonstrate a time trend. However, 92% of adults had a comorbid condition. Comorbid conditions were much more common with adult hospitalizations compared with pediatric hospitalizations (92% vs 63%). The mean LOS for adults with comorbid conditions was 98% longer than the mean LOS for adults without comorbid conditions. Longer LOS was seen for adults with comorbid conditions for all 12 conditions examined.

The proportion of adult hospitalizations exceeding the M&R LOS criteria is greater than that for pediatric hospitalizations for all 12 conditions examined. The extent to which LOS exceeds the M&R guidelines, especially relative to LOS for children, can be clearly seen in Table 4. In most cases, the percentage of hospital stays exceeding the guidelines is far greater for adults, with the largest difference occurring for hospital stays for pneumonia, for which 75% exceeded the guidelines compared with 29% for children. The percentage of bed days exceeding M&R LOS criteria is greater for adults than for children for all 12 diagnoses. The largest difference occurs with hospital stays for pneumonia, with total bed days exceeding the M&R guidelines by 51% for adults, whereas total bed days exceeded the guidelines by 25% for children.

With the growing use of case-rate payments and capitated contracts for pediatric care, much more attention will be paid to LOS and related guidelines introduced by M&R,⁵ among others. While hospital discharge data from Pennsylvania show that there are a significant number of bed days that exceed the M&R LOS criteria for pediatric conditions, it seems that the extent to which LOS exceeds M&R LOS criteria is considerably less than what was found using discharge data from New York State.⁴ Sills and coauthors found that 13 of the 16 conditions examined had median LOS that exceeded the M&R functional LOS guidelines. In fact, there were several instances when M&R LOS criteria exceeded modal LOS in our study. There are several reasons why this divergence may have occurred. One possibility is that there are notable differences in practice styles in New York and Pennsylvania. Past analyses have shown significant geographical variations in medical practice for identical conditions.^10,11 It is also possible that there are significant differences in the case mix between the 2 states. In Pennsylvania, 63% of children had more than 1 diagnosis recorded. It is possible that an even greater percentage of children in New York experienced medical complications associated with comorbid conditions. As LOS were greater for adults and adults were more likely to have comorbid medical conditions than children, it seems that comorbidity is associated with increased LOS. The extent of managed care penetration or other characteristics of the health care market (eg, total hospital beds available) may also account for differences in observed LOS between the 2 states. Finally, the Pennsylvania data are for hospitalizations occurring in 1996 through 1998, while the report by Sills and colleagues used data from 1995. An increase in pressures to contain costs during this period or other secular changes may explain some of the observed differences. This last explanation seems unlikely as there was no observable time trend in the Pennsylvania data between 1996 and 1998.

Although this study found fewer conditions for which the modal and median LOS exceeded the M&R LOS criteria than did the study by Sills et al,⁴ between 29% and 83% of hospital stays and between 20% and 64% of bed days exceeded the M&R LOS criteria. Pediatric institutions would be placed at financial risk if case rates and contracts were based on M&R criteria without considering the numerous pediatric patients with hospital stays likely to exceed the M&R LOS criteria. We also observed several instances in which modal and median LOSs were less than the M&R LOS criteria. Overall, it seems that M&R LOS criteria are not consistent with routine practice, with functional M&R LOS criteria being potentially too short in some cases and potentially too long in others.

There are methodological issues that encourage caution in the interpretation of our findings. First, the data are drawn from the Pennsylvania Health Care Cost Containment Council hospital discharge database, an administrative dataset. As with all administrative datasets, coding of diagnoses is not perfectly accurate, and coding of diagnoses may be variable across sites. It is not clear how such a bias would affect results, unless it was not distributed randomly. Also, such a database cannot adequately describe the outcomes for patients with early or late hospitalizations such as those with iatrogenic illness, rehospitalizations, or mortality. To determine a causal relationship of LOS guidelines on child health, a prospective study would be essential.

Despite these limitations, there are several implications for guideline creators and users from this study and prior work. First, it is important for those producing guidelines like M&R to use the best possible science (eg, evidence for routine practice) in the production of LOS guidelines. The M&R guidelines are meant to represent best practice,⁵ but they are likely applied as routine practice.⁶ It seems that the M&R guidelines may need to be revised to represent routine practice, with criteria falling both above and below median and modal LOS for many conditions. It is concerning that for at least 2 conditions (bacterial meningitis and osteomyelitis), both our study and that by Sills et al⁴ found the guidelines to be inconsistent with routine practice from 1995 through 1998.

In addition, those using the guidelines as payers or providers should understand their purpose—that is, to act as an ideal timeline in the absence of serious complicating issues or comorbid disorders. In our study, 63% of all children and 92% of adults had at least 1 comorbid diagnosis listed. Such comorbidity is likely to be at least in part responsible for the numerous children (more than one fourth) who exceeded guideline-recommended LOS even for those conditions in which the median did not.

Accepted for publication March 16, 2001.

The views expressed in this article do not necessarily represent the views of the University of Pittsburgh.

We gratefully acknowledge the assistance provided by the Marketing Department of Children's Hospital of Pittsburgh, and we thank Jeffrey Whittle, MD, for his helpful comments on this manuscript.

Corresponding author and reprints: Jeffrey S. Harman, PhD, University of Pittsburgh School of Medicine, 3811 O'Hara St, Suite 430, Pittsburgh, PA 15213 (e-mail: harmanjs@msx.upmc.edu).