Economic incentives and diagnostic coding in a public health care system

A number of countries have introduced activity-based payment systems for hospital care by linking all or part of the hospital budget to the number of discharged patients while at the same time adjusting for treatment intensity or patient complexity (case mix). The diagnosis-related group (DRG) is one of the most common systems used to account for case mix. DRGs are widely used for both monitoring and payment purposes. The size of the reimbursement differs between patients, reflecting differences in complexity and thus treatment costs. Patients are categorized in different groups based on diagnosis and procedural codes routinely registered in medical records. For some groups, the DRG system makes the distinction between a “complicated” and an “uncomplicated” patient. While the main diagnosis will be the same, complicated patients will have one or more additional “complicating” secondary diagnoses. Within the resulting pair of DRGs, the complicated group will thus have higher predicted costs and a higher reimbursement. Because personnel in hospitals register information about diagnosis, there is the possibility that a patient is consciously coded to a “complicated” DRG. This is often referred to as “upcoding” or “DRG creep”, first defined as “a deliberate and systematic shift in a hospital’s reported case mix in order to improve reimbursement” (Simborg 1981). It has also been argued that the introduction of activity-based payment systems will increase the importance of accuracy and completeness in coding (Fisher et al. 1992; O’Reilly et al. 2012). The latter view is shared by the Norwegian government body responsible for the Norwegian DRG system, which defines DRG creep as “patients being coded as more complete, resulting in an increase in case mix index” (translated by the authors from Helsedirektoratet (2011)). Indeed, evidence from the US Medicare system indicated that the introduction of a prospective payment system in 1983 was followed by an increase in the average case mix (Carter and Ginsburg 1985; Ellis and McGuire 1986; Carter et al. 1990; Stern and Epstein 1985; Rosenberg 2001).

In the past decade, there has been a renewed interest in issues related to DRG creep and upcoding. Examining a policy reform in the financing of US Medicare discharges, (Dafny 2005) found a positive association between price differences between complicated and uncomplicated DRGs and the share of discharges in complicated groups. More recently, Barros and Braun (2016) found a positive association between price incentives and upcoding in Portugal.

Responses to price incentives vary between different types of hospitals. In Sweden, the increase in the number of secondary diagnoses registered was larger in hospitals with prospective payment systems than hospitals without prospective payment systems (Serdén et al. 2003). Two studies in the USA found that for-profit hospitals were more likely than non-profit or government-owned hospitals to upcode (Dafny and Dranove 2009; Silverman and Skinner 2004), and also that hospitals in “economic distress” were more likely to upcode (Silverman and Skinner 2004). However, no difference in upcoding between public and private hospitals was found in Italy (Berta et al. 2010).

In a cross-country comparative study, Steinbusch et al. suggest that health systems combining for-profit hospitals with the use of secondary diagnosis criteria for classification, such as in the USA, were more susceptible to upcoding (Steinbusch et al. 2007). In a systematic review, Palmer et al. argued that the effects seen in other countries are similar to those observed in the US system (Palmer et al. 2014). In a theoretical work, Kuhn and Siciliani suggested that the level of auditing of the financing system will influence the perceived risk related to upcoding, and this can also explain differences in levels of upcoding across health systems (Kuhn and Siciliani 2008).

The purpose of this paper is to add to the relatively small literature on upcoding in systems dominated by public hospitals by providing an analysis of coding behaviour in Norway over a period of 10 years. The Norwegian health care system is tax funded, with universal access to services that are largely free at the point of use. Hospitals are predominantly publicly owned and financed through a combination of global budgets and activity-based funding. Activity-based financing was introduced in 1997 utilizing a Nordic version of the DRG system. In the period covered by this study (1999–2008), the share of activity-based funding fluctuated between 40 and 60 %.¹ The period also encompasses a major ownership reform in 2002, where hospital ownership was transferred from 19 county councils to the state (Magnussen et al. 2007).

Analysing coding behaviour in the Norwegian health care sector allowed us to address three questions. First, in a public health care system, the additional income generated from upcoding remains in the hospital. Thus, it will be used to increase the level of activity beyond what was planned, to increase slack (inefficiencies), or it will be saved to finance future investments. It remains uncertain to what extent actors in this public setting will seek to increase income by upcoding. Second, the substantial changes in the degree of activity-based funding during the period studied allowed us to analyse to what extent public hospitals adjust their coding behaviour in response to changes in financial incentives. Third, using observations over a period of 10 years allowed us to study any underlying trends in coding behaviour, and isolate this from the effects of changes in financial incentives. In all three questions, our main interest was the potential relationship between economic incentives and coding behaviour on an aggregate national level. Although there are numerous micro-level examples of upcoding (Lægreid and Neby 2012; Neby et al. 2015), it is unclear whether these are exceptions to the rule, or whether they represent a general behavioural response to economic incentives.

Our goal was to examine the association between the potential gain in income from upcoding and the coding behaviour of hospitals. Across DRG pairs, we found a positive association between the gain in income from upcoding and the share of discharges classified as complicated. Thus, DRG pairs in which there was a higher gain in income from upcoding also had a higher share of complicated discharges. However, although we controlled for co-morbidity, age and length of stay, we cannot exclude the possibility that this partly reflects differences in the case mix. Nevertheless, it is not clear why the difference in treatment costs between complicated and uncomplicated discharges should be higher in DRG pairs with a higher share of complicated discharges and therefore our results indicate that coding behaviour is related to the size of the incentive.

We found that a difference in price between a complicated and uncomplicated group of one DRG point was related to a difference of 14 percentage points in the share of complicated discharges within a DRG pair. Although this may seem like a large effect, the average potential gain from upcoding was only 0.28 DRG points (see Table 2).

We found no association between changes in \(\Delta p_{it}\) over time and the share of complicated discharges within a DRG pair. Thus, in a period with frequent changes in the share of activity-based funding, hospitals did not seem to respond by changing their coding behaviour. However, when stratifying the analysis by medical and surgical DRGs, we found a small, positive association for medical DRGs. Because surgical patients are generally more homogeneous (within a DRG) than medical patients, there may have been less opportunity for tactical coding of these patients. Although the size of the estimated association was small, this result indicated that there might be subgroups of patients where the relationship between financial incentives and tactical coding is stronger. This corresponds to earlier results on how Norwegian hospitals respond to price changes (Januleviciute et al. 2016). Melberg et al. have recently shown higher growth in DRG groups with a price increase than in groups with a reduction in reimbursement rates (Melberg et al. 2016).

We found that the share of complicated discharges increased during the ten year period covered by the study. This may be due to changes in case mix resulting from demographic changes, changes in technology, changes in the quality and completeness of coding and finally changes in the financing system. Recalling the two different definitions of upcoding and DRG creep presented in the introduction, we cannot here distinguish between “deliberate upcoding” and “more complete coding”. The increasing trend could both indicate that the quality of coding has improved, and at the same time that the presence of explicit and implicit incentives is followed by a general increase in the recording of secondary diagnoses. Thus, while we cannot label all upcoding as being completely driven by financial incentives, we argue that such incentives were present and that their consequences are reflected on an aggregate level by the increasing time trend. The introduction of activity-based funding in 1997 was followed by an increased use of secondary diagnoses. Eventually the use of secondary diagnoses will reach a level (or equilibrium) where it might be difficult to justify an additional secondary diagnosis from a medical point of view. Thus, one might suspect that a large part of the potential for increase was exhausted in the period following the hospital reform, explaining the slowing growth in the share of complicated discharges.

This paper decomposed the price incentive into two components, \(\overline{p_i}\) and \(\Delta p_{it}\), to differentiate between the level and changes of the incentive for upcoding. This approach differs from earlier studies but demonstrates that, in Norway, the differences in prices are more important than changes within groups. Hospitals may appear to respond to prices, but the changes in price are probably too small to have a large-scale impact.

We believe that the major strength of this analysis is the fact that we are able to utilize a complete dataset covering all DRG pairs for all patients at all hospitals. Our analyses include a ten year period in which there have been large and repeated changes in the potential gain in income from upcoding. Thus, any aggregate effects of increased gain in income from upcoding should be detected in this study. By controlling for a time trend and separating within and between effects, we are more reassured that any remaining effects are more related to upcoding rather than to an increase in the quality of coding.

We have employed a system perspective by pooling all DRG pairs, hospitals and years in the same analysis. This could dilute important findings for specific DRG pairs. Silverman and Skinner (2004) found substantial evidence of upcoding for patients with pneumonia. Their results were robust to different model specifications, but sensitive to the included DRGs. Our stratification showed very different results for the medical and surgical DRG pairs. It is safe to assume that even larger differences will be found on examination of separate DRGs. However, our aim was to detect system-level effects and not effects of singular groups or hospitals. One might also question whether the observed changes in the price incentive were large enough to have an effect. While frequent and potentially substantial, the changes in incentives observed in this study were small compared with some of the larger exogenous shocks described by, for example, Dafny (2005). Therefore, it may have been unrealistic to expect significant results from the observed changes. A change of 20 percentage points in the share of activity-based funding is, however, not trivial and it is interesting that these changes only seem to have led to a marginal change in coding practice.

Upcoding can take place in all systems that incentivize documenting of diagnoses. We have limited our study to upcoding in DRG pairs in Norway. These groups amount to less than one-third of the total volume of treatment. Upcoding is possible for all groups, but the paired structure of complicated/uncomplicated lends itself easily to our research strategy of testing directly whether incentives are associated with upcoding. There are several ways “manipulations” can occur in a DRG system (Neby et al. 2015). In this paper, we have focused solely on upcoding and not touched upon other related strategies: gaming, dumping, skimping and skimming. Further studies should attempt to distinguish upcoding from other manipulations empirically. It is impossible using registry data to determine whether the upcoding has been deliberate. To assess the actual conscious decision to upcode, one must opt for a qualitative approach. This study has not ventured into the auditing of diagnosis and hospital records. Earlier evidence from Norway has indicated that diagnostic accuracy is not very high (Jørgenvåg 2005), and it would be interesting to consider whether the Norwegian auditing scheme could be considered optimal (Kuhn and Siciliani 2008).