Evolution of case-mix in heart surgery: from mortality risk to complication risk

Abstract

During the last two decades despite an increase of the average preoperative mortality risk of patients referred to heart surgery a decrease of hospital mortality has been observed in many surgical institutions. The ratio between the increase of risk and the decrease of mortality could be defined as the ‘risk paradox’ for coronary surgery. Meanwhile an increase of the incidence of postoperative complications is leading to a longer stay in intensive care that involves a remarkable cost increase per single hospitalisation and a disproportionally long-term use of reanimation beds in those patients who survive the operation but have comorbidities complicating the postoperative course. This progressive change of the epidemiology of patients undergoing heart surgery is coupled with a progressive increase of costs. In the present review a comparison of stratification models developed to predict hospital mortality with those developed to predict prolonged stay in intensive care is discussed. Such predictions are not obviously aimed at deciding whether to operate a patient or not, but can be looked in managing high risk patients, e.g. by a daily monitoring and revision of their prognosis and relevant therapeutic choices, as well as in discussing with their relatives about whether to continue or not implacable treatments. After identifying the models, it is desirable that they are spread into professional Societies in order to sensitise field operators’ awareness on the issue of proper intervention indications and on the opportunity of identifying those patients for whom an intervention is not to be advised and to whom propose medical or intervention treatments.

1 Mortality risk stratification

The belief that for a quality evaluation of intervention results it is necessary to stratify patients according to the risk of adverse events has become evident within the scientific community of heart surgeons and Health administrators of the western world over the last 20 years. Even though, up to the 1990s the predictive models were developed only to predict the hospital mortality risk; subsequently in the last 10 years, new models have been published enabling us to stratify patients according to unfavourable postoperative course, characterised by complications, long hospital stays, high costs, as well as poor results in terms of quality of life.

All risk stratification models (there are over 20 models, even though less than ten of them are extensively used) [1] have been developed by selecting the factors significantly associated with different types of risk out of several dozens of demographic, clinical and angiographic variables. In some models, such as the American model from the Society of Thoracic Surgeons[2], the variables contained in the register used to update the stratification model are over 100; in the other more famous models, such as those by Parsonnet [3], of the State of New York [4] and the more recent Euroscore model [5], the variables entered into the registers exceed 40.

As long as clinical and epidemiological scenarios and heart surgery technology are evolving, the possibility of discovering a significant association among the preoperative characteristics of the patients, quality heart surgeons, and quality of cardiac surgery units, as well as the short and middle term outcomes, is linked to the availability of databases complete and specific for the type of surgery, regularly checked and updated. In fact, when, on the same population, a standard risk model, a risk recalibrated model and a risk model especially updated with new risk factors were compared based on the characteristics of the operated population [6], it has been demonstrated that the application of a standard model may provide an inaccurate assessment of the performance of either Heart Surgery Units or single heart surgeons.

2 Increase of risk increase and decrease of mortality

Nowadays, the evaluation of data available [7–18] has certainly proved that the average mortality risk of patients undergoing heart surgery has been gradually increasing for at least 20 years [9,19,20]. For instance, when the Parsonnet model was applied to the surgical population of the Beth Israel Hospital in Newark [9], it was observed that the average risk of bypass patients gradually increased from 6.5 (in 1988) to 9.6 (in 1994), while, in the population of the National Adult Cardiac Surgical Database Report[20], it increased from 5.0 (in 1996) to 6.0 (in 1999), and at the Cleveland Clinic Foundation[19] from 2.9 (in 1996) to 3.1 (in 1999). In fact, over the last 20 years, patients undergoing bypass surgery have been older (average age increased from 58 to 65 years), with lower ejection fraction (a reduction from 62 to 49%), with more severe coronary heart disease, and a higher proportion of patients with left main disease (an increase from 7 to 20%). They were operated in medical emergency conditions (increasing from 5 to 40%), and with major associated comorbidities (for instance, the rate of patients affected by renal failure increased from 2.5 to 4.2%).

On the other hand, wide surveys in the USA [19,21] and in the United Kingdom [20] pointed out that, in contrast with the increase of risk as evaluated through different models, hospital mortality decreased during the same period; from the national database of the Society of Thoracic Surgeons, reduction of mortality from 3.8 (in 1991) to 2.6% (in 2000) has been observed, whereas, during the same period, at the Cleveland Clinic Foundation, the reduction was from 6 to 1.8%, and, from the data of the Society of Cardiothoracic Surgeons in the United Kingdom and Ireland, from 2.4 to 2.1%.

The gap between increase of risk and decrease of hospital mortality is statistically expressed by the fact that risk prediction models tend, in recent years, to overestimate expected mortality [22,23] (Figs. 1 and 2) , mainly in patients belonging to higher risk classes [20]. In fact, from the Society of Thoracic Surgeons's experience published by Ferguson [22] it was observed that, based on data obtained from risk preoperative prediction, expected mortality is lower than observed mortality up to 1994, observed and expected mortality are equal in 1995, whereas, on the contrary, an overestimation is observed in subsequent years. Similarly, in the State of New Jersey's experience [23], expected mortality is lower than observed mortality up to 1995, but it becomes higher since 1997. If we take account of the gap between expected and observed mortality based on patient risk class, we notice that, as the preoperative risk increases, and also the overestimation obtained from the stratification model increases.

The inverse relationship between the increase of risk and decrease of mortality could be defined as the ‘risk paradox’ of coronary surgery. Different hypotheses were devised to explain this phenomenon: the technological progress in surgery, anaesthesia, and reanimation procedures (general use of arterial grafts, warm cardioplegia, improvement in ventilation and nutritional techniques), is probably the main cause of the decrease in observed versus expected mortality [7,16–18] also for those patients who were considered at high risk in the past. In recent years, it has been realised that the burden of risk factors in the models developed 10–15 years ago, reflected a surgical technology unable to make high risk patients survive, who today do not die during the intraoperative period. For instance, if the regular updates of the model developed by the Society of Thoracic Surgeons[17] are sequentially examined, we will discover that, for some risk factors, the odds ratio (i.e. the index of the relative weight of a single factor versus mortality) is changed; an increase of the odds ratio shows a greater impact of that risk factor in explaining the event. During the last few years, the odds ratio of some factors (re-intervention, emergency intervention, ejection fraction, preoperative renal failure) decreased; therefore, the weight of these factors has decreased. In Edwards’ opinion [17], this trend will lead to a condition where single preoperative risk factors will play an ever smaller part in predicting operative mortality, and death will be more and more often considered as an event not linked to the severity of patient's condition, but to an accidental, unpredictable event.

3 Mortality and complications

According to the majority of authors, the risk factors of hospital mortality are the same as the risk factors of complications [11,24–27]; this means they can induce both mortality and complications. Christakis's analysis [7], based on data obtained from Toronto University cardiac surgery units between 1982 and 1986, showed that, against a decrease of mortality from 12 to 6%, there was an increase of non-fatal complications from 10 to 13%. Similarly, comparing 1981–1987 with 1993–1995 [8], the decrease of mortality and the increase of some non-fatal complications (neurological, pulmonary and mechanical complications needing an intra-aortic balloon pump) have shown an opposite trend, whereas the incidence of wound infection decreased (Fig. 3) . Then, it can be suggested that a number of patients who would have died during past years, at present do survive despite several postoperative events, which complicated their hospital course and prolong their hospital stay.

From the analysis of published surveys, it is not always easy to understand the shift from mortality to complications. In fact, the proportion of complicated patients is very low in the majority of low risk patients; even a significant increase of complications in the few higher risk patients is diluted by the evaluation of the overall population and may not be supported by a statistically significant result, because mortality and complications are usually referred to the overall population, not to patients belonging to different risk classes. Consequently, it cannot be estimated if the incidence of adverse events significantly increases in the patient groups who would have been affected by a high mortality in the past.

4 Increase of the length of stay

During the last decade a similar phenomenon has been observed in the evaluation of trends in length of stay in intensive care and in hospital. Actually, many authors consider long postoperative length of stay as an event correlated to complications [20,24,28–30], to such an extent that it is considered a surrogate end point.

According to the Society of Thoracic Surgeons’ survey, the hospital length of stay has gradually decreased since 1991, both on average (from 6.8 to 4.9 days) and as median (from 9.9 to 6.8 days) [31] (Fig. 4) . In this case too, the decrease of the length of stay is evaluated on the whole population and thus influenced by the overwhelming prevalence of cases with a short hospital course. From the overall analysis of the data, it cannot be observed if few patients are forced to a much longer stay than during past years. In fact recently, for many non-complicated patients the fast track discharge program has been adopted, so further decreasing the mean and median stay of the entire population of operated patients.

These data, observed in the USA, have not been confirmed by the National Adult Cardiac Surgery Data Base Report's survey [32], collected in the UK. Here an increase of the average postoperative course, mainly for elderly patients, for patients with re-intervention, and with a severe ventricular dysfunction was observed. The mismatch between the USA and UK, rather than an index of a different type of patients or a different incidence of postoperative complications, is probably due to the different reimbursement method of cardiac surgery units in the two countries. In a national health system such as the British one, where a rather limited flat rate budget is annually set for every unit, the cardiac surgery units have no advantage in performing more than pre-set interventions. In fact, they would not have any revenue increase. In many cases, an extension of the hospital course neither matches the patient's real needs nor the types of complications occurring during hospitalisation. On the contrary, in a system such as the American one, where financing is linked to the number of procedures carried out, a strong incentive is introduced to decrease the length of stay in intensive care, in order to increase the number of surgical interventions.

In some studies, patients forced into a prolonged stay in intensive care and hospital were evaluated to determine their destiny in terms of follow-up, quality of life and resources consumption. The proportion of patients who have a prolonged course in intensive care is around 5–6%, whereas the proportion of patients with prolonged hospital course ranges from 8 to 37%. This gap is not only due to the severity of operated patients’ conditions or to the experience of individual surgeons. It can also be due to the cut-off used to define ‘prolonged stay’ (from 3 to 10 days in intensive care and from 7 to 14 days in a postoperative course, according to single studies) [33–36] (see Table 1) .

Nevertheless, the distribution of the length of stay in intensive care and hospital is very similar in different studies and shows an asymmetric curve, characterised by a shift of the mean and median to lower values and by a long tail of less frequent values [27,35].

5 Destiny of patients with a prolonged stay in intensive care

The survival and quality of life of patients with a prolonged hospitalisation in intensive care are particularly low (see Table 2) .

In Ryan's study [37], hospital mortality is 42% for patients (3.8% of the whole population) who remained in intensive care for more than 14 days. In Bashour's study [27], the hospital mortality of patients who remained in intensive care unit for more than 10 days (5.4% of the whole population) was 33%, compared to 1.5% of patients who had been transferred earlier than 10 days (94.6% of the whole population). Holmes's data [38], show that hospital mortality is 33.3% for patients hospitalised in intensive care unit for over 48 h (7.2% of the whole population). In our experience, by increasing the threshold of length of stay in intensive care, we observed that the rate of patients decreases (21.3% were hospitalised for over 48 h, 6.1% for over 5 days and 3.3% for over 10 days) and hospital and follow-up mortality increases (0.8% for patients hospitalised for less than 48 h, 12.5% for hospitalisations longer than 48 h, 36.6% for hospitalisations longer than 5 days and 50% for hospitalisations longer than 10 days). For surviving patients, a greater incidence of complications, which involves a prolonged hospitalisation, affects post-discharge quality of life; 2 months after coronary surgery, the Duke Activity Status Index's median value on the 50 patients who were discharged after 10 days intensive care was 26 against 58 expected value [27].

6 Resources consumption by patients with a prolonged stay in intensive care unit

The prolongation of a patient's stay in intensive care involves a remarkable increase in costs per single hospitalisation and a disproportionately high use of reanimation beds for a long period of time versus non-complicated patients. Lahey [29] calculated the average cost per bypass patient in a 924 patient population and he noted that the average stay cost for the patients who were hospitalised less than 14 days (period obtained for the hospitalisation of patients with Diagnosis Related Groups 106) totals $29 903, whereas the patients hospitalised for a longer period of time doubled it, up to $59 177. Even though patients hospitalised for over 14 days are only 10% of the population, they cost, on the whole, more than the remaining 90% of patients, hospitalised for less than 14 days ($11 480 000 against $10 735 000). In Bashour's study [27], it is shown that 5.4% of the patients hospitalised in intensive care unit for over 10 days consume almost half of the total resources used in an intensive care unit and 23% of the overall resources spent for the postoperative stay. If patients are classified according to the occurrence of complications during the postoperative period [25], it results that 48% of patients has no complications and uses 32% of resources, 40% has at least one minor complication but no major complications and uses 41% of resources and 12% has at least one major complication and consumes 27% of resources.

Bed occupancy is also highly conditioned by patients with a prolonged stay. According to Tu [35], 73% of patients in intensive care for less than 3 days, used 23% of beds, whereas 27% of remaining patients use 77% of beds. So, patients who are in intensive care for over 10 days (only 6.2% of operated patients), use almost half (47.2%) of intensive care beds. In an elderly population, 90 patients who died in hospital (8%) used 30% of beds in intensive care and the 625 complicated patients (54%) used 86% of beds.

In our experience, a direct cost analysis performed only on the population of 288 patients who underwent coronary artery bypass surgery demonstrated that 55 patients (19%) who remained in intensive care for over 2 days consumed 30.4% of resources, whereas 233 patients (81%) with less than 2 days course consumed 69.6% of resources.

7 Risk factors of prolongation of stay in intensive care and in hospital

Considering the unfavourable course and the disproportionately high consumption of resources by the patients with a long stay in intensive care, research into preoperative and intraoperative risk factors predicting the risk of complications and the length of stay well deserves the attention of heart surgeons and Health administrators. Some authors demonstrated a correlation between Parsonnet mortality risk scores [11,20,39] or other models [24,29] and the duration of stay in hospital or intensive care (Figs. 5 and 6) . Other authors [25–28,30,33–37] preferred to develop original models after proving the poor accuracy of mortality risk models to the purpose of predicting the length of stay. If clinical and hemodynamic characteristics of patients are entered into predictive models at the admission in intensive care [40] or after a number of hours [38,41,42], the predictive value of mortality of patients with a prolonged hospitalisation improves up to the identification of 97% of patients destined to die [38]. This is true to a greater extent if these characteristics are integrated with preoperative characteristics [28,40].

Identifying the risk factors of a prolonged stay is not of course an argument against intervention; it has been known for a long time and more recently confirmed [43] that patients at increased risk due to the severity of heart conditions, expect higher survival rates from surgery than from medical treatment. However, this is unlikely when the high risk is due to a non-cardiac co-morbidity, such as a neurological, renal or pulmonary disease. For instance, in our experience, patients with high cardiac risk (ejection fraction <30%) had a preoperative risk calculated with the Euroscore equal to 8 and a hospital mortality equal to 7%; however, patients with preoperative co-morbidities and similar Euroscore to the one of patients with reduced ejection fraction had much higher hospital mortality. These data confirm the need to select the patients to be operated, not only on the grounds of an overall risk evaluation, but also on the grounds of the type of factors contributing to determine the preoperative risk of mortality and length of stay.

Obviously, such predictions are not only aimed at deciding whether to operate a patient or not, but they are also useful in managing high risk patients, e.g. by a daily monitoring and revision of their prognosis and relevant therapeutic alternatives, as well as in discussing with their relatives whether to continue or not heroic treatments.

A comparative analysis of the models of preoperative mortality risk and of the models for the prediction of the length of stay risk showed that the former had a greater homogeneity than considered variables [1]. With the exception of the Parsonnet model, validated in three Units [11,24,39] and used also on British [20] and Euroscore register data, quite recently used by the authors who helped its development [44], none of the other models specifically developed for predicting length of stay has been exported and validated in units other than those where they were created.

8 Final considerations

Heart surgery is at a turning point. The observation of data contained in national and multinational registers shows that since the mid 1980’s the following changes have occurred:

• an increase of average preoperative risk of mortality and complications;

• an increase of the percentage of interventions on patients with a high risk of mortality;

• a decrease of operative mortality (risk paradox);

• an increase of the incidence of postoperative complications;

• a high mortality in patients with a prolonged stay in intensive care;

• an increase of costs.

The recurrence of these phenomena involves difficult ethical, professional, economic and organisational problems. For the majority of low risk patients, heart surgery, and in particular coronary surgery, can nowadays be done with a high safety profile. On the contrary, for high mortality risk patients, the death that during past years could occur in the operative room, may nowadays occur after a long stay in intensive care or a few weeks and months later after discharge. However, this is not only an issue of mortality; in fact, the treatment of these patients is often dramatic for the patients themselves, owing to the physical and psychological suffering they are exposed to, for their relatives, who are subjected to a very stressful time [45], and for staff, due to hard work, and frustration after frequent failures in spite of their maximum efforts. In addition, patients subject to a prolonged postoperative stay may occupy the intensive care beds for many days, and need a high usage of supplies (blood and hemoderivates, drugs, mechanical ventilation, etc.) and the performance of procedures in excess (tracheostomy, intra-aortic balloon pump, etc.) which, besides affecting the rhythm of the operating theatre, are turned into a cost that sometimes is equal to the cost for all other much more numerous patients with non-complicated courses.

Considering the great significance that a knowledge of the long stay in intensive care and hospital population still can have in clinical, organisational and economic decisions, it is reasonable to carry out researches aimed at comparing already validated models to the purpose of checking the ones which are more accurate in predicting the prolongation of stay in intensive care. After identifying the models, it is desirable that they are spread to professional Societies in order to sensitise field operators’ awareness on the issue of proper intervention indications and on the opportunity of identifying those patients for whom an intervention is not to be recommended and to whom medical or angioplasty treatments should be suggested.

References