The Road Ahead for Healthcare in England

  • Print
  • Connect
  • Email
  • Facebook
  • Twitter
  • LinkedIn
  • Google+
By Simon Moody | 07 September 2011

How to manage the risk within a tighter budget?

One of the most important issues in the current discussion of reforms in the English National Health Service (NHS) concerns budget accountability. Healthcare costs are not exempt from the government’s public spending cuts, and it is widely understood that stricter accountability is needed to achieve financial results without compromising quality of care. The focus of accountability is shifting from the existing primary care trusts (PCTs) to general practitioner (GP) practices or, perhaps more likely, to Clinical Commissioning Groups (CCGs).

Central to the issue of greater accountability is the question of whether and how to develop a person-based resource allocation formula that would draw upon risk-adjustment models to allocate budgets for GP practices and/or CCGs, which would then be responsible for their specific budgets. Risk adjustment (RA) is not used in England to the extent that it is in other European countries such as Germany and Holland, but it does have applicability within the NHS. The approach to resource allocation in England has utilised age, gender, and (to some extent) geographical factors, but it has never drilled down to the level of patient diagnoses or other details of the population’s medical history.

This article considers some of the factors in the potential application of an RA model to healthcare budget allocation in England.

Why a risk-adjustment approach?

The NHS budget is going to be squeezed in the near future as the British government tightens the nation’s belt. Future expenditure rises will have to be controlled as the health service seeks to save £15 to £20 billion by 2014. The Department of Health (DH) does not want to sacrifice the quality of care that Britons have come to expect from their system, and so the savings should, in theory, come mostly through increasing the efficiency of healthcare delivery. This is not necessarily a bad thing for patients; done right, efficient healthcare delivery can also mean quality healthcare.

The assumption behind the ideas now under discussion is that the people who know best how to manage care are those on the front line; and the people at the front of the front line are the GPs, those who first see a patient (except in emergency room cases). So the argument for reform is to give them the budget; they know best what their patients need. And that should improve the quality of care as well.

This is where risk adjustment can become an important component of the allocation process. At its best, RA is an effective tool for identifying high-risk, and therefore, high-cost patients within a given population and for adjusting healthcare budgets according to risk levels. Thus, the allocation for a given practice or CCG would be calculated by a formula based on the combined financial risk of its population.

For financial accountability to work, the process needs to have some teeth; for example, bonus payments for living within the budget while maintaining high standards of care and/or financial penalties for failure to do so. Otherwise there will be little incentive for budget holders to improve efficiency and the quality of care. The way the system is structured now, with budgets channelled through the PCTs, the process is more bark than bite. The amount by which an underperforming PCT exceeds its budget is supplemented from another PCT that has performed well; if none are performing well, ultimately more funding has to come from the central taxation pot.

Adding teeth to the process should give GP practices and CCGs a strong incentive to examine their expected volatility in light of their budgets and take pains to manage their budgets carefully. This will not be a popular measure, but it is necessary in order to keep costs under control.

The risks of risk adjustment

There are some problems with RA models. In the first place, experience from the United States shows that most prospective models have an overall predictive accuracy no greater than the R-squared value1 of 30%. Concurrent or retrospective models generally score higher, but rarely higher than 60%. Given these figures, one might well question the models’ ability to explain differences in healthcare expenditures effectively. However, RA models are always more accurate than the present models that are based on global age and gender analysis. RA has been improving and can provide many useful insights with better design, richer data, and more advanced modelling techniques.

Prospective and retrospective risk adjustment

The use of RA for calculating equalisation payments in insurance systems can yield some lessons for applying RA to budget allocation within the National Health Service. Equalisation payments can be calculated either prospectively (also called ex ante) at the beginning of a particular period or retrospectively (also called ex post or concurrent), during or following the period. A combination of the two approaches is also common, whereby equalisation payments are first calculated prospectively and then adjusted retrospectively.

Calculating budget allocations prospectively means setting allocations at the beginning of a budget period based on global risk-adjustment factors such as age, gender, and geography, as well as historic healthcare service utilisation or diagnostic information aggregated from previous periods. In contrast, retrospective models would utilise data arising during the period for which the RA is calculated. As with equalisation payments, it would be possible to apply a prospective RA model for budget allocation and make adjustments for its inaccuracies retrospectively.

To the extent that RA has been used in England, it has had some application to budget allocation and predictive modelling in the sense of risk stratification, i.e., identifying the high-cost individuals, but it has not been used retrospectively.

Any budget allocation process utilising RA must take into account the challenges posed by both prospective and retrospective models. Prospective models place relatively more weight on systematic factors such as ageing and chronic conditions rather than acute and one-time conditions. Therefore, prospective models can create incentives to avoid and not over-diagnose acute episodes of care. This poses a potential moral hazard problem with respect to the many health conditions for which treatment or prevention activities are discretionary. In addition, as one can infer from the R-squared data, no matter how good the ex ante adjustments, actual costs incurred will differ from projected costs.

Retrospective models generate their own challenges. Reimbursement based on actual cost reduces the incentive for organisations to provide efficient care or to invest heavily in medical management, because the financial benefits of any efficiency improvements will be mostly shared with other organisations through the retrospective adjustment. Without efficiency incentives, overall health expenditures will be driven up. Alternatively, reimbursement might be based on diagnosis codes, compensating provider organisations according to average cost of treatment. That, however, would deprive organisations of a financial incentive to provide care beyond the cost of the average reimbursement, putting the sickest patients at risk of not receiving all the care they need.

In all likelihood, the solution will be a blend of prospective and retrospective modelling.

Implementation challenges. Implementing RA involves numerous issues besides the choice of prospective and/or retrospective models. The following are a few that have important consequences for budget allocation:

  • Choice of risk factors to include: Age, gender, and diagnosis and/or pharmacy information2. But also geography? Socioeconomic factors?
  • For the initial calibration, are the appropriate data sets available and do they represent the current environment? Similarly, when recalibrating the model, does the data set show the most recent past experience, and have any clinical coding classifications changed?
  • Have physicians coded their diagnoses accurately, specifically, and consistently? Are changes in risk scores due to true increases in the illness burden, or to changes in coding? Are diagnoses being reported more complicated than they actually are? And are changes in coding practice due to changes in incentives?
  • In the NHS, patient attribution should be straightforward when patients are registered with a GP, a CCG, or a PCT, but what about those who are not registered? Also, how should the system handle the attribution of new patients and patients who move organisations during the fiscal year?
  • How frequently should risk-adjusted budget allocation take place? It may not be administratively practical to make adjustments more often than once a year, even though risk-bearing organisations (especially those with higher-cost patients) might want more frequent budget allocation.

Transparency. Finally, risk adjustment raises questions of transparency. The NHS commissioning budget is approximately £80 billion, and how this budget is dispersed to CCGs and then GP practices is a matter of utmost consequence, not only to those who administer the funds and provide healthcare services but, ultimately, to the millions of patients served. However, most RA models and methodologies are proprietary and not readily available to users; the process can easily look like a black box, and if stakeholders don’t understand what’s going on inside the box they most likely will have less trust in what comes out of the box.

Ideally, the RA system should provide direct insights into what medical conditions contribute to the risk score, and by how much. When combined with evidence-based medicine algorithms, this information can be useful to those responsible for medical management, as it can help identify areas for clinical intervention and estimate cost-saving opportunities.

With this in mind, one shouldn’t neglect the fact that clinicians are key stakeholders, so clinical credibility is another important consideration. Regardless of whether it affects the predictive accuracy of the RA model, if clinicians observe large differences in payments based on apparently trivial classification differences, this will undermine the clinicians’ acceptance of the process.

Size matters

Because RA models are far from perfect, there will always be variabilities in healthcare expenditures that the models cannot explain. This puts smaller provider organisations at particular risk. Smaller practices, and even smaller CCG consortia, face a double whammy. First, they have greater underlying risk volatility—i.e., they are especially vulnerable to exceeding their risk-adjusted financial allocation because of high-cost patients and events. Second, because they have less data to feed into the RA calculations, the predictive modelling underlying their budget allocation is going to be less accurate than that of larger groups.

Thus, the need to think in terms of organisational consolidation. There’s a lot of talk today about the formation of Clinical Commissioning Groups, what their size should be, how many GP practices might come together to form a consortium. Most of the discussions to date have focused on the minimum size necessary to manage healthcare efficiently in terms of administration and overhead, rather than the minimum size necessary to successfully manage financial risk.

Three options. There would appear to be three broad options for finding a solution to the problem posed by small practices’ ability to offer quality care within their budget allocations:

  • Ring-fencing all high-risk procedures or diagnoses—i.e., stripping out the high-cost cases from a provider’s budget and having someone else manage the risk. High-risk, chronic conditions, mental healthcare, high-cost maternity cases and all other risky, essentially one-off acute cases—these would not be a part of the organisation’s allocated budget but would, instead, be managed by specialists within the system. The GP organisation’s risk is thus limited to the more manageable and predictable cases.

    The ring-fencing approach, however, diverges from DH’s goal of assigning GPs the accountability for all, or most, of the care of their populations, so it is not clear that this solution is viable.
  • Risk pooling. There is already a form of unstructured risk pooling in the NHS; if a practice exceeds its budget, then someone else will pick up that overspend and the first party will meet its budget. There’s a sense in which that doesn’t seem fair and equitable, because there don’t appear to be any accurate calculations of the risk that a particular party brings to the pool, i.e., the risk that it will exceed its budget. Rather like the problems of retrospective risk equalisation mentioned earlier, there is limited incentive to improve the efficiency of care provided if the resulting gain is then distributed to those who are less efficient.

    If the NHS turns to risk pooling in a regular, structured way—which seems the most likely scenario, given that that’s the informal arrangement at the moment—there will need to be some form of risk assessment and a premium attached to the arrangement. Such a plan might resemble a traditional insurance, or self-insurance, arrangement, in which GP practices and CCGs pool their risk and share in each other’s fortunes or losses.
  • A third option might be for practices and CCGs to consider arrangements with private reinsurers to cover their solvency through stop-loss or defined lump-sum payments to cover the financial risk of costly one-time events. This approach has the advantage of bringing external capital into the system, but it is fraught with complications both financial and political. First, it requires reinsurers willing to insure the healthcare organisations involved. The reinsurers would naturally want to make a profit, and this raises the potential that reinsurers will want to have a say in the conditions of care—a prospect that is distasteful to many NHS stakeholders.

In conclusion

Clearly, there is a connection between budgets allocated via risk adjustment and the volatility of a GP practice’s risk. In other words, the smaller the GP practice, the more likely they are to experience greater risk volatility and the more difficult it will likely be to stay within their allocated budgets.

Thus, it makes sense for smaller practices to form CCGs with other practices in relation to the management of financial risk. Such consolidation might be difficult in certain circumstances, and even small CCGs are likely to experience risk volatility beyond their ability to make their budgets. How can smaller organisations control, transfer, or share their risk? That is the question as the NHS moves forward with budget-allocation reform.

Provider organisations need to start thinking about insurance-style solutions to managing risk, particularly the risk of incurring financial insolvency. Actuaries can play a role here. Actuaries have generally not been involved extensively in managing financial risk in the NHS; nearly all have been active only in the private insurance sector. But the new challenges facing the NHS will require the kind of skills actuaries can offer. Actuaries are experts in finding insurance-style solutions to the challenge of managing risk.

Of course, the word “insurance” is not a popular expression in the NHS context. Nevertheless, provider organisations need to start looking into risk-sharing and/or risk-transferring solutions. None have been developed yet, but it is time to pay attention to the impending challenge.

In every healthcare system, there two drivers: cost and quality. The NHS will have to draw a fine balance between how much accountability there can be and how much it’s safe for the government to allow. But ultimately, we live in times when there is less budget to go around. Provider organisations will have to manage those budgets much more tightly while still delivering quality healthcare to their patients.

1 R-square is an individual-level measure of how much variability in the total healthcare expenditure can be explained or predicted by the model. A perfect model has an R-squared value of 1.0, or 100%. A model with no explanatory power has an R-squared value of 0.0, or 0%.

2 Diagnosis-based RA models are generally preferred over pharma-based models because they usually have more intuitive results and higher predictive accuracy. However, in the absence of complete diagnosis coding, pharma-based models may be preferable as long as drug codes are standard, timely, and of good quality.