OUP user menu

Exploring the impact of complications on length of stay in major surgery diagnosis-related groups

Julián Librero , Manuel Marín , Salvador Peiró , Alberto Verdaguer Munujos
DOI: http://dx.doi.org/10.1093/intqhc/mzh008 51-57 First published online: 4 February 2004


Study objective. To analyze, in terms of the length of stay (LOS), the use of resources by patients classified under surgical diagnosis-related groups (DRGs) with complication and/or comorbidity (DRGCCs), divided into subgroups where complications were and were not detected, and to explore the repercussions on hospital reimbursement.

Design. The Complication Screening Program© (CSP) was used to divide the patients in 14 DRGCCs into subgroups with and without complications, and to compare the LOS between the subgroups and with the equivalent DRG without CC. The ratios between the LOS and the relative weight of the DRGs for reimbursement were also compared.

Setting. The population hospitalized for major surgery between 1995 and 1999 in the Hospital of Mataró, Catalonia, Spain.

Participants. Patients (4227) hospitalized for major surgery.

Main results. The percentage of complications identified by the CSP ranged from 17.5% to 52.4%. The LOS of the DRGCCs was almost twice as long as in the DRGs without CC; 2.48 times greater if the DRGCC was flagged by the CSP and 70% greater if it was not. On average, the DRGCCs selected were reimbursed at a rate 93% higher than their counterpart DRGs without CC. In 11 of the 14 DRG pairs with or without CC, the reimbursement ratio was lower than the LOS ratio.

Conclusions. DRGCCs can be classified into clearly differentiated groups based on the presence or absence of CSP-flagged complications. CSP-flagged complications produce an increase in LOS greater than the increase in the relative weights for reimbursement.

  • administrative data
  • diagnosis-related-groups
  • hospital complications
  • length of stay
  • quality of care

Diagnosis-related groups (DRGs) are used widely as a tool to classify patients into groups on the basis of the resources they are expected to consume during hospitalization. These groups serve to establish the hospital’s budget or revenues, and can also help to examine the financial implications of different clinical management strategies. DRGs have been an important instrument in the shift from retrospective payment systems, based on reimbursement per stay or historical budgets, to prospective payment systems, which provide incentives to improve hospital efficiency. If the payer reimburses a fixed amount per case (the average cost of a given DRG), the hospital can keep the difference between the costs of less expensive hospitalizations and the average cost of the DRG, but will incur losses if hospitalization costs are higher [1,2].

When the DRGs were built, length of stay (LOS) was used as the outcome variable because it provides a strong correlation with costs [3]. Cases were initially distributed in large, mutually exclusive diagnostic groups called major diagnostic categories (MDC). Each MDC was then separated into DRGs on the basis of the values of specific variables (main and secondary diagnoses, age, surgery procedures, and reason for discharge) according to a statistical algorithm that could later be modified to maintain clinical sense. The identification of certain secondary diagnoses, which are specific to each DRG, will change a case from a DRG without comorbidities and/or complications (DRG w/o CC) to its counterpart DRG with CC (DRGCC), thus implying greater relative weight for reimbursement. The type of DRGs used in this study—Health Care Financing Administration (HCFA) Grouper 16—includes 511 DRGs, of which 228 are pairs with or without CC.

When a given DRG warrants reclassification to its counterpart with CC, no distinction is made between a secondary diagnosis of a condition that was present when the patient was admitted to the hospital (comorbidity), or one that developed during the hospitalization (complication). In fact, there are understandable difficulties in distinguishing between comorbidity and complications on the basis of the information provided on the Minimum Basic Data Set (MBDS) at hospital discharge [4], but this confusion may have implications for the incentives to provide better or worse quality care. In fact, although it is reasonable to reimburse the costs derived from the greater comorbidity of the patients, it is not clear that reimbursement should be made for costs derived from avoidable complications.

Although in-hospital complications can be expected from the natural course of illness or may arise as adverse secondary effects of appropriate treatments, they may also be related to the greater or lesser quality of the care provided [5,6]. Under this last assumption, the inability of the DRG system to differentiate between illnesses coexisting at the time of admission and that developed during the patient’s stay may provide, in theory, disincentives to improve quality if the marginal profits of reimbursement received when the patient goes from one DRG w/o CC to its counterpart with CC are higher than the additional costs related to the complication. The aim of this study is to analyze, in terms of LOS, the resources consumed by the patients in specific surgical DRGCCs, with and without complications, and to explore the repercussions of this use of resources on hospital reimbursement schemes.

Materials and methods


We used the Complication Screening Program© (CSP) to separate specific surgical DRGCCs into subgroups ‘with’ and ‘without’ complications, comparing the LOS between the two groups, and with those corresponding to the counterpart DRG w/o CC.


We reviewed summaries of the computerized discharge records on the MBDS of the Hospital of Mataró, a center in the Public Hospital Network of the Catalonia Health Service for the County of Maresme, Catalonia, Spain. Since 1997 the Catalonia Health Service has used a reimbursement system based, in part, on the DRG classification system. The Mataró hospital is among the most active Spanish hospitals in monitoring diagnostic coding, with an average of 4.3 diagnoses recorded per episode since the implementation of the DRG payment system, a figure that is well above the average for Spain (fewer than three diagnoses per episode).

Selection of patients and DRGs

Between 1 January 1995 and 30 June 1999 the Mataró hospital admitted for 6929 patients hospitalization in many of the 92DRGs included in the ‘major surgery’ group of the CSP. These are surgical DRGs that do not include the terms ‘minor’, ‘other’, or ‘miscellaneous’, where the average length of the stay is >3 days. Of the total, 54 DRGs (n = 5194) corresponded to pairs with and without CC. We selected 28DRGs (14 pairs, n = 4227), which included >10 cases in the three pre-defined analysis subgroups: DRG without CC (DRG w/o CC), DRG with CC but without CSP complications (DRGCC CSP-unflagged) and DRG with CC and CSP complications (DRGCC CSP-flagged).

Identification of complications

The version of the CSP used in this study was based on the report outlining the original version of the program [7], which has been described in detail in various papers [810]. The CSP was designed to identify potentially preventable complications on the basis of information contained in administrative databases [age, sex, codes from the International Classification of Diseases 9 Clinical Modification (ICD9CM) and LOS from admission to intervention], using the hospitalization episode as the unit of observation. Twenty-six algorithms, built with specific ICD9CM codes for secondary diagnoses or procedures (trigger codes), identify complications in major surgery. When these are detected, they suggest that a complication may have developed (i.e. pneumonia by aspiration). Upon detection, these codes trigger a series of computerized algorithms that verify whether the case meets specific conditions defined for each situation. For example, should a secondary code for pneumonia by aspiration (ICD9CM 507.0) be detected in a patient undergoing major surgery, the algorithm will confirm that the patient presents neither epilepsy, trauma, drug overdose, or poisoning as the main diagnosis; if none of these conditions are confirmed, it will identify the case as a possible post-surgery complication.


First, the LOS and the percentage of complications in each of the selected DRGs were described. The next step looked at the DRGCC and compared the LOS of the cases that were flagged as complicated by the CSP with those that were unflagged, or where no complications were detected; the LOS of the latter group were then compared with those of the DRGs w/o CC. Student’s t-test was used for independent samples assuming no homogeneous variances. A one-sided test was used to test the hypothesis that the LOS would be greater in the subgroup of CSP-flagged DRGCCs compared with CSP-unflagged DRGCCs, and that the unflagged DRGCCs would show longer LOS than the DRG w/o CC. Statistical significance was established at a P-value of <0.05. Finally, the ratios between the LOS for the different subgroups were calculated, as were those for the relative weights (HCFA fiscal year 2000) of the cases with and without CC in order to compare the relative increases in LOS with the relative increases in weight for reimbursement. A ratio value of 1 indicated that the stays in the subgroups compared were of equal length. The statistical analysis was performed using STATA© software Version 5.


Table 1 shows the pairs of DRGs selected for analysis and their relative weights, together with the LOS and the percentage of complications identified by the CSP. The average LOS for all cases was 9.2 days, with a range from 3.6 (appendectomy without CC) to 25.5 days (pancreas, liver, and shunt procedures with CC). The LOS was always greater in the DRGs with CC than in their respective pairs without CC. The CSP only flagged two cases in the DRGs w/o CC (one in DRG 167 and the other in DRG 359). For the DRGs with CC, the percentage of cases flagged as complicated by the CSP ranged from 17.5% (DRG 197, cholecystectomy except by laparoscope with CC) to 52.4% (DRG 150, peritoneal adhesiolysis with CC).

View this table:
Table 1

Relative weights, number of cases, LOS, and percentages of CSP complications in selected pairs of major surgical DRGs

Table 2 reports the average LOS in the DRGs with CC that were both flagged and unflagged as complicated by the CSP (18.6 and 11.4 days for flagged and unflagged cases, respectively), together with the average LOS for the DRG w/o CC (6.7 days). The subgroup of DRGCCs flagged as complicated by the CSP always presented greater LOS than the unflagged group, although in DRG 110 the difference between subgroups was not significant. The subgroup of DRGCCs not flagged by the CSP showed consistently longer stays compared with the respective DRG w/o CC (greater LOS in 13 out of 14 of the pairs), although in three of these pairs (110–111, 146–147, 150–151) the differences were not significant. In one case alone (DRGs 257–258), the DRG w/o CC presented an average LOS that was slightly greater than its counterpart with CC; however this difference was not significant.

View this table:
Table 2

Length of stay by DRGCC, with and without complications, and by DRG w/o CC in selected pairs of major surgical DRGs

Table 3 shows the relationship between increases in LOS (proxy of costs) associated with the presence or absence of complications identified by the CSP, and increases in the relative weight (proxy of reimbursement). The first column shows that the DRGCC group (whether flagged or not by the CSP) showed LOS that were, on average, close to twice as long as those of the DRGs w/o CC and, if we apply the USA weights (second column), they would receive reimbursements that were 93% higher than the corresponding pair without CC. Once the DRGCCs had been separated into two subgroups depending on whether the CSP had identified complications or not, the unflagged DRGCCs showed LOS that were 1.70 times higher that the DRGs w/o CC (third column), while the flagged DRGCCs had LOS that were 2.48 times higher than the DRGs w/o CC (fourth column). Finally, when comparing the DRGCCs with and without complications flagged by the CSP (fifth column), the flagged DRGCCs stayed 46% more days compared with the unflagged DRGCCs. In 11 of the 14pairs, the ratio of relative weights between the DRGCCs and the DRGs w/o CC (second column) was lower than the ratio of the LOS between the DRGCCs flagged by the CSP and the DRGs w/o CC. Pairs 110–111, 148–149, and 154–155 were the exception, and showed the opposite. In the subgroup of unflagged DRGCCs, the ratio between the relative weights of the DRGCCs and the DRGs w/o CC was always greater (except in the pair 166–167) than the ratio between the LOS when the DRGCC subgroup was compared with the subgroup of DRGs w/o CC.

View this table:
Table 3

Ratios between length of stay or relative weights in function of the type of DRG and the presence of complications


DRGs and their relative weights have been used in the United States to introduce a prospective payment system in Medicare. This system has proven its effectiveness in containing hospitalization costs [11,12], although not necessarily in containing overall health care expenditures. Evaluations conducted to date have not registered relevant changes in terms of the quality of care provided [1315]. These factors have contributed to the adoption of the DRGs in Europe, where they enjoy widespread acceptance [16]. However, because of the many models in use and health care reforms currently underway, in Europe DRGs are used for many different purposes, from total or partial reimbursement to their simple application to adjust comparisons between the LOS in different hospitals (profiling), without any financial implications in this instance [17].

The bulk of criticisms lodged against the DRGs over the last two decades concentrate on their inability to incorporate legitimate differences in costs between different groups of patients [1821], which results in the penalization of hospitals with patients presenting greater severity. In theory, these differences should be detectable with more sophisticated risks adjustment systems [2226], although in practice most of these systems do not improve the explanation of the variance in costs of the classic DRG system [2729] and have, to a large extent, also failed to differentiate between comorbidity and complications.

It is, nonetheless, of the utmost importance to be able to discern between comorbidity and complications from the perspective of the quality of the care provided. All health care interventions carry risks that are not always foreseeable or avoidable. However, ample evidence also demonstrates that there are ways of preventing complications by improving surgery techniques, fine-tuning the evaluation of risks, increasing patient follow-up, and, in general, improving care provision processes. Clearly there is a difference between a patient who is admitted with a stroke accompanied by pneumonia, and another who is initially admitted for stroke, but who develops pneumonia during the hospital stay. By the same token, and positing that there may be a basal level of complications even in hospitals that provide optimum quality care, it would be one thing to see a complications rate of 2% and quite another to see a rate of 20%. Hence, with this distinction in mind, it would seem logical to reimburse additional costs incurred in treating a patient with comorbidity and, even, to reimburse for a basal rate of complications. However, it would be counterproductive to reimburse additional costs derived from an excessively high rate of complications due to inadequacies in the care provided.

The rate of cases flagged as complicated by the CSP in this study (11.5%) is similar to the rates of 9.7% [9] and 10.8 % [30] detected in databases from California, although in these cases, all the major surgery population was taken into account, making no distinction between the different DRGs. When analyzing costs, Kalish and coworkers found that after adjusting for specific covariables, cases with complications showed costs that were 97% greater than cases without complications, although these authors did not classify the cases as DRGCC and DRG w/o CC [30]. Munoz and colleagues analyzed different surgical and medical DRGs w/o CC, and in both the classical DRGs and the ‘all patient’ DRGs, they found that cases with complications or comorbidity had higher costs. These authors, however, did not stipulate a specific DRGCC classification, nor did they not differentiate between comorbidity and complications [31,32].

Various factors limit studies like the one reported here. The first is the use of the average length of stay as the unit to measure costs. Although this variable is known to have an excellent correlation with costs [3], discrepancies may arise with the incorporation of costs, such as the doctors’ fees and the price of prostheses, which may not have been taken into account in the calculation of the relative weights of the DRGs. Furthermore, in the case of surgery, it is more probable that the relationship between the LOS and the costs incurred is not linear, since the bulk of the costs are related to the surgical intervention itself [30] and the longer or shorter length of the post-operatory stay does not correspond to proportional increments in costs for the institution. Similarly, it is difficult to evaluate the cases separately on the basis of the LOS: in some cases the stay is prolonged because of a complication; in others the long stay is propitious for the development of a complication, since there seems to be a basal risk of developing a complication associated with the stay itself [33,34]; and in a third group, prolongation of LOS increases the probability of registering complications that, despite developing simultaneously, would not be incorporated in the MBDS as they occurred after discharge [35,36], or that, in all events, would be registered elsewhere. It should be pointed out, however, that studies based on more detailed clinical information suggest that prolongations in LOS are mainly due to complications [37]. Discussions are even focussing on the possibility of determining a threshold in the LOS beyond which the proportion of patients with complications would be significantly higher [38].

The second limiting factor of this study is associated with the limitations of the CSP itself in differentiating between comorbidity and complications, and between unavoidable complications and complications due to quality problems. The CSP was designed to be used with administrative databases, and its authors have pointed out that it is better used as a triage instrument [810] than as a means to detect quality problems associated with individual hospitalization episodes. In an early validation exercise that used explicit criteria to judge processes, a clear correlation has not been found between problems related to the care provision process and the observation of complications flagged by the CSP [39]. A later study evaluating the application of the CSP in major surgery cases found that it had positive prediction value that was >80% [40]. In a further study conducted together with the previous one as part of a common validation project in major surgery, the reviewers identified potential quality problems in 29.5% of the cases flagged by the CSP [41]. In the Spanish context, the positive prediction value of the CSP was 90%, although the instrument was extremely sensitive to the quality of the data available [42], underlining a further limitation that lies with the highly questionable quality of the information in the administrative databases like the MBDS [4346].

Other problems inherent in a study of this nature are associated with the limitations of the ICD9CM itself, and with the aspect of obtaining data retrospectively. This study may be particularly affected if the complications were registered more often with increased length of the stay. This bias is plausible, since it is likely that the greater the severity of the complications (more hospitalization days) the more thoroughly they were registered. Similar biases have been demonstrated in the MBDS with the registration of comorbidity [35,46,47].

Two final limitations affecting this study are the low number of cases in some of the analyses, and the fact that confidence intervals were not used in the comparison between the ratios. We opted not to use them because the relative weights represented a pre-fixed value.

In conclusion, bearing in mind the limitations outlined above, the results of this study underscore the fact that DRGCCs flagged as complicated by the CSP are associated with longer stays than the unflagged episodes. These results should not be seen as a recommendation for basing reimbursement on CSP complications, a policy that could have the negative effect of removing incentives to reduce the rate of complications. To the contrary, and while the limitations of the CSP in identifying complications related to quality problems make it impossible to draw definitive conclusions, the results of this study suggest that, at least in the context of major surgery, the DGRs contain incentives for providers to make an effort to reduce surgical complications. This aspect reinforces the development of strategies to improve quality in order to maximize profits.


We would like to thank Dr Lisa I. Iezzoni, who provided the documents for reconstructing the Complication Screening Program© and granted permission to use it in this study. This project was funded by the Fondo de Investigación Sanitaria (FIS 99/0888), with complementary aid from the Fundación Instituto de Investigación en Servicios de Salud.


View Abstract