SEARCH

SEARCH BY CITATION

Summary

  1. Top of page
  2. Summary
  3. Competing interests
  4. References
  5. Appendix

Mortality is a good measure of killing, but it is a poor measure of cure, palliation or the maintenance of function. Nevertheless, it has remained the primary metric of hospital care for 200 years. This article discusses the factors that contribute to mortality risk and survival trajectories, as well as the increasing recognition that surgery kills for months after the last suture is tied. This article discusses how disparate factors can usefully combine to generate an ‘elderly’ group with a monthly mortality in excess of 1% and a median life expectancy less than 3.5 years. A downloadable spreadsheet is provided that combines risk factors to generate mortality risks and their associated survival curves, emphasising the importance of looking beyond one postoperative month.

If the function of a hospital were to kill the sick, statistical comparisons of this nature would be admissible.— Florence Nightingale, 1859 [1]

Mortality is an insensitive measure of care: most patients survive the postoperative period, whilst (perhaps) many observed deaths are not associated with poor care. We should concentrate on the living, not the dead. Unfortunately, information on the effect surgery has on people's lives is scarce, in comparison with how often it kills them. This paper discusses how to estimate the chance that someone will die and how much this might change following surgery, but in so doing, we might reflect on Florence Nightingale's observation that mortality is an excellent measure of performance only for those trying to kill.

Chance (at least in part) determines survival and death, which can be estimated using certain characteristics throughout adulthood: age; sex; historical morbidity; and physiological function [2-12]. The latter two characteristics are meant in their broadest sense, so that together they might accommodate variables that are, or might be, independently associated with survival. For instance, socioeconomic status is an historical characteristic that confers morbidity and mortality risks, whereas brain natriuretic peptide (BNP) is a marker of physiological function that might be shown to have an independent prognostic value. Crucially, variables used in the calculation should have a known independent effect on long-term survival in the general, non-surgical population, so that all the purposes of pre-operative mortality estimation are served (see below). Variables, such as BNP, have yet to declare their value in promoting the precision of peri-operative prognostication, because their relationship with general survival has been inadequately described.

It is one's risk of death, rather than one's proximity to birth, that determines the trajectory of one's remaining days. The ‘elderly’ might therefore be better characterised by a certain hazard of dying per week or per month than by how long ago they were born. For instance, rather than define ‘elderly’ as older than 65 years of age or 75 years of age etc, one might better define the concept as, for example, a median survival less than 5 years or a monthly mortality greater than 1 in 100. This approach liberates the categorisation of ‘elderly’ from being synonymous with age alone. However, adults with quite disparate characteristics (whether age, co-morbidity etc), might form a common group with equal numerical mortality risk and survival prospects. Paradoxically, this approach, defining ‘elderly’ as a risk rather than an age, allows one to incorporate age into the calculation without fear of ‘ageism’: one should discriminate, but on the basis of ‘riskism’. Nevertheless, age constitutes the largest risk factor for most adults.

One's mortality hazard – the risk of dying this month – is inextricably linked with one's survival trajectory, as long as the hazard is not temporarily increased by acute changes, for instance, riding a motorcycle at high speed or mounting a physiological response to surgery. If we know someone's current mortality hazard, we can plot the trajectory of their survival over subsequent years; conversely, if we know a parameter of their survival trajectory, for instance, their median life expectancy, we can determine their current mortality hazard (Fig. 1) [13].

image

Figure 1. Decreasing median life expectancy (vertical axis, years) as monthly mortality rate increases (horizontal axis, %). The axes are respectively crossed at two values that one could use to define ‘elderly’: a monthly risk of 1.0% (median life expectancy 3.5 years), or a median life expectancy of 5 years (0.8% monthly risk). The line is generated from the UK Interim Life tables 2009–11 [13].

Download figure to PowerPoint

The horizontal axis of monthly mortality risk, rather than age, not only has the advantage of accommodating the common effect of disparate risk factors, but will also accommodate the year-on-year fall in mortality rates that populations (in developed countries) have experienced over the last 50 years. In 2009, for example, the age at which the ‘average’ man and woman registered a 1 in 100 monthly mortality rate in the UK was 86 and 89 years, respectively, compared with 80 and 85 years, respectively, in 1981. A median life expectancy of 5 years was reached 1–2 years sooner in both cohorts [13].

The pre-operative estimation of postoperative mortality and morbidity has at least three purposes. The purpose familiar to most anaesthetists is to target peri-operative interventions to reduce both morbidity and mortality. The assumption common to all interventions – whether pre-operative, operative or postoperative – is that they will have a larger absolute effect on patients who have a higher mortality risk. This assumption is wrong in detail, which might be important on occasion. The opportunity for preventing death in two patients with the same risk of dying in the first postoperative month might be very different: for one, this risk might be the same as his/her normal risk, if he/she were having minor surgery for example, whilst for the other, this risk might be ten times his/her normal risk. One might reasonably suppose that it is the change in mortality risk that better matches the value of interventions and, on this basis, one might determine the second purpose of risk estimation better, that is using data to distribute resources more rationally.

The third purpose of risk estimation is to create a context within which patients can determine their own fates through collaborative decision-making. Unfortunately, this purpose is usually confused with the first purpose (targeting interventions towards reducing mortality and morbidity). For instance, the NCEPOD report Knowing the Risk: a Review of the Peri-Operative Care of Surgical Patients predominantly views risk from a clinician's perspective rather than that of a patient – the mortality rate 1 month after an operation has almost no value for the patient (see below) [14]. A patient has to make a choice whether to proceed with an operation, balancing the chances of harm and benefit with each choice. After the vast majority of scheduled procedures, the mortality rate 1 month later will exceed the rate without surgery, hence how does mortality information help the patient other than to suggest that no one should have scheduled surgery?

Over the last 200 years, medicine has generally failed to record postoperative outcomes of value to patients. Instead, in the UK, the middle of the nineteenth century witnessed a confluence of various factors that catalysed the publication of hospital mortality statistics, including the appointment of a Registrar General, the standardisation of death certification, the classification of diseases, the alignment of city hospitals with Universities and scientific discipline, and the import by Florence Nightingale of policies that had reduced hospital mortality in the Crimean war [15].

Exactly the same problems that perplex the interpretation of mortality statistics nowadays also perplexed the Victorians:

accurate hospital statistics are much more rare than is generally imagined, and at the best they only give the mortality which has taken place in the hospitals, and take no cognisance of those cases which are discharged in a hopeless condition, to die immediately afterwards, a practice which is followed to a much greater extent by some hospitals than by others.” [1, 16]

Mortality rates are dependent on what one uses for the numerator and denominator, what method one uses for risk adjustment and how thoroughly one checks whether risk factors are present or not. The papers investigating variation in mortality rate between hospitals paint a picture of a compromised metric, which reflects the assiduity of clerical staff, the validity of databases and the machinations of those in a position to bias the calculation [17-22].

Assessment of death at 30 days was an attempt to avoid bias caused by hospitals discharging moribund patients. A follow-up period of one synodic lunar orbit (29.53 days) is undoubtedly arbitrary, but at least it is fixed. It is also longer than most hospital admissions and therefore more likely to discriminate between different mortality rates. For instance, the 30-day mortalities for myocardial infarction, pneumonia and heart failure are twice the in-hospital mortality rates [19]. Traditionally, the performance of hospitals in the USA has been measured with in-hospital mortality, which has favoured hospitals with shorter lengths of stay because, unsurprisingly, more patients die the longer you watch. Shorter hospital stays have generally shifted deaths from the hospital to the community. The mean length of hospital stay for heart failure, for example, fell from 8.6 to 6.4 days between 1994 and 2006, with the in-hospital death rate falling from 8.2% to 4.5% and the community death rate rising from 4.4% to 6.3% (to 30 days after admission) [23]. The European Surgical Outcomes Study (EuSOS), the only large, prospective, international, epidemiological study of unselected surgical patients, used in-hospital mortality as the primary outcome, which reached 3% for elective surgery, 5% for urgent surgery and 10% for emergency surgery [24]. In that study, 60-day mortality was at least twice the 30-day postoperative rate. EuSOS is not the only study to suggest that elective surgery can increase mortality for at least 60 postoperative days. Thirty-day mortality after major surgery in 105 951 US patients was also 3%, after which mortality appeared to be elevated for another 1–5 months across a range of operations [25]. The interpretation that surgery elevated postoperative mortality beyond 30 days was based upon ‘inflection’ points in the survival curve. Visual inspection of survival curves is an appealing technique, but it fails to quantify the magnitude to which surgery increases mortality, as well as being susceptible to subjective interpretation. The other feature of that study was the importance attributed to postoperative complications as a determinant of subsequent mortality, both at 30 days and beyond. In another study, the excess mortality caused by injury took up to 6 months to resolve, although 90% of that excess was exhausted within 60 days of injury [26].

The introduction of anaesthesia in 1846 … increased the pressure on the wards, straining buildings designed for a less active style of medicine.” [15]

It is evident from both randomised trials and observational data that scheduled surgery increases mortality, sometimes for more than 3 months, with the postoperative survival curve taking more than 1 year to equal the survival of patients who had not had surgery. Survival and quality of life, months and years after surgery, are therefore the measures of the success, the failure and the wisdom of operating in response to surgical pathology. By (my) definitions, the elderly do not have many years of life left to lose, with a median life expectancy of 3.5 or 5 years (Fig. 1), but are at considerable risk of high postoperative mortality and lost function and lost independence, without gain.

The elderly also focus the mind on how to distribute scarce peri-operative resources. What metric should be used to prioritise limited peri-operative resources to patients – years of life saved, quality-adjusted years of life saved, absolute risk of dying, absolute increase in risk of dying caused by surgery, or some combination of these? What levels of peri-operative care are available and what metric thresholds could one use to determine their provision? Who will decide to postpone surgery when the allotted level of care is unavailable, or to continue regardless – the surgeon, the anaesthetist, the geriatrician, the intensivist, the patient, his/her family, or some combination of these? Is critical care, or any other peri-operative intervention, more able to limit a postoperative monthly mortality of 2% generated by an 8-fold increase of a 0.25% pre-operative risk, or 3.5% mortality generated by a twofold increase of a 1.75% pre-operative risk, even though both scenarios have an absolute increase in risk of 1.75%? Or would the mortality be affected to the same extent in both scenarios? Should critical care be preferentially provided to people with ‘more life to gain’, even if their absolute postoperative risk of dying is less than those denied it? The elderly will, again by (my) definition, always have a higher absolute mortality risk, with any given operation generating a greater absolute increase in mortality, than someone who is not elderly. Hence, to put it the other way, should one deny the ‘young’ critical care in preference to someone with a greater absolute increase in risk, but a smaller relative increase in risk, that is the elderly?

Quantifying pre-operative mortality: determining ‘who is elderly’

If you subscribe to the view that ‘elderly’ can best be defined by a given mortality rate or life expectancy, then you will need a calculator to estimate these. I have developed a spreadsheet (https://www.dropbox.com/sh/i57xhrtyop0pq31/tLGppuJGWs) that will generate monthly mortality rates and survival curves, using the following sequence (see Appendix Appendix for details):

  1. Average survival, according to UK Interim Life Tables [13] (http://www.ons.gov.uk/ons/publications/re-reference-tables.html?edition=tcm%3A77-274529), which provides the average mortality for a person the same age and sex as your patient.
  2. Co-morbidities [2-12]. The calculator then adjusts the average survival figure for each of six diagnoses (myocardial infarction (MI), angina, stroke, transient ischaemic attack, heart failure, peripheral arterial disease).
  3. Renal function [27-29]. The calculator then adjusts for estimated glomerular filtration rate.
  4. Physical fitness [30-32, 12, 33-41]. The final adjustment is for physical fitness.

This sequence calculates a monthly mortality, framed as a risk of ‘1 death per n people per month’, with the calculator reporting the value of n. A survival curve is then generated by following the reduction in population expected from the closest Interim Life Table figures to the calculated mortality rate, but adjusted for an assumed future 1.4% annual fall in relative mortality.

A second mortality rate and accompanying survival curve can also be calculated, adjusting for body mass index (BMI), such that the underweight are penalised, and the overweight (to a point) ‘rewarded’. This reflects evidence from the general population (including the elderly) and from peri-operative studies that the overweight have a lower mortality than both underweight (BMI < 20 kg.m−2) and ‘normal’ weight patients (BMI 20–25 kg.m−2) [42-46].

Quantifying postoperative mortality: the effect of surgery

The most reliable way to determine the effect of surgery on survival is to conduct a randomised controlled trial. In the ‘UK Small Aneurysm Study’, the mortality rate in the first month after open abdominal aortic aneurysm (AAA) repair was 12 times that in the surveillance group, hence surgery temporarily increased mortality 12 times [47]. One can also attempt to interpret uncontrolled data by generating expected mortality rates, calculated from the variables listed above, then comparing them to observed rates. In the EVAR 1 study, mortality one month after open repair was ten times that expected, reassuringly similar to that found in the UK Small Aneurysm Study [48], compared with a rate of 4–5 times that expected one month after EVAR.

Respective multipliers of 10–12 and 4–5 for open and endovascular AAA repair respectively can be used as benchmarks around which the monthly postoperative risks of other procedures can be estimated. For instance, one can estimate the expected postoperative mortality rates for the range of 3.7 million elective operations reported retrospectively by Noordzij et al. [49]. Following open AAA repair, the observed-to-expected mortality ratio was 24, or about double that measured in the UK Small Aneurysm Study/EVAR 1, suggesting miscalibration, perhaps due to miscoding of procedures or incomplete reporting of patient characteristics. Estimating the effects of non-AAA repair surgery, therefore, requires observed/expected mortality ratios from Nordzij et al.'s paper to be decreased by half.

Figure 2 shows the postoperative survival curve (solid black line) generated for a patient considering whether or not to have an elective, open AAA repair. The curve is generated after imputing a mortality rate increased 10-fold compared with baseline in the first postoperative month. After this point, survival is determined by modifying the survival curve expected for a similar patient who had neither surgery nor AAA (dashed line). This (falsely) discounts an elevation in mortality beyond one postoperative month, even though one would expect mortality to remain increased up to 3 months or more after open AAA repair. However, survival bias operates to counteract this effect; in the aftermath of major surgery, the diminished population would be expected to be ‘fitter’ than an unexposed, unselected population, as a result of older, frailer, sicker members of the exposed population having died from the trauma of surgery. Survivor populations usually demonstrate less steep survival trajectories, therefore, than control populations. The shaded area in Fig. 2 corresponds to expected survival without surgery, bordered by estimates of slow and fast AAA expansion (entered into the calculator).

image

Figure 2. Survival curves generated by the calculator described for a 75-year-old man (172 cm, 65 kg), previous MI, eGFR 55, who has a 5.5 cm diameter abdominal aortic aneurysm (AAA) in 2013. Curves are with open AAA repair (black) or without surgery (grey zone). Both curves would benefit from a third axis, quality of life.

Download figure to PowerPoint

Risk stratification in practice

The explicit quantification of mortality affords clinicians and patients a tool that helps collaborative decision-making and distribution of peri-operative resources. Perhaps 5% of the UK population would be classified as ‘elderly’ using a definition of monthly mortality more than 1%. Elective total hip replacement in an elderly patient will increase absolute mortality in the first postoperative month by at least 1%, with more ‘risky’ operations generating commensurately higher mortality rates.

Torbay Hospital has explicitly quantified, recorded and communicated mortality estimates for patients having a range of operations for more than 6 years, the monthly mortality figure holding little informative value for patients compared with their calculated survival curve. However, the monthly figure helps to guide estimates of the risk of survival accompanied by prolonged disability, loss of function and loss of independence, and is used to trigger escalation in peri-operative vigilance, planning and preparation. The hospital has been trialling a risk model for inpatient surgery with four strata, based upon mortality in the first postoperative month of < 1%, 1–3%, 3–6% and > 7%. The hypothesis is that increased intensity of peri-operative care, for instance involving postoperative critical care, can reduce both mortality and morbidity when postoperative mortality exceeds 1% [50, 51]. We have further data (unpublished) that support these tentative conclusions: over a 3-year period, 311 patients with colorectal cancer had their postoperative mortality estimated, 129 of whom had a risk 1–3%. We observed that none of the 37 who had planned admissions to intensive care subsequently required unplanned intensive care admission. Conversely, 13 out of 92 who were discharged to a general surgical ward from recovery subsequently required urgent admission to intensive care, predominantly due to anastomotic leaks.

These data have made clinicians cautious about proceeding with scheduled operations that increase mortality risk by more than 1% if appropriate systems of care are unavailable, which has increased the number of operations postponed due to lack of critical care facilities. This suggests that calculating mortality risk might have ramifications for healthcare as the numbers of older, frail patients increases, because it exposes care insufficiencies that can no longer be overlooked.

The observation quoted from Rivett [15], that Victorian hospitals were unable to cope with the increase in patients afforded by the development of anaesthesia, is unfortunately prescient. Hospitals, particularly critical care systems, were not designed to cope with the increasing flux of elderly surgical patients. By calculating peri-operative mortality risk, it is possible to quantify inadequacies in the provision of peri-operative care, and act to ameliorate any further decline in these as the demand for ‘elderly’ care increases over the longer term.

Competing interests

  1. Top of page
  2. Summary
  3. Competing interests
  4. References
  5. Appendix

No funding or competing interests declared.

References

  1. Top of page
  2. Summary
  3. Competing interests
  4. References
  5. Appendix

Appendix

  1. Top of page
  2. Summary
  3. Competing interests
  4. References
  5. Appendix
  1. The Office for National Statistics (ONS) publishes mortality data for the United Kingdom. The numerator is the number of deaths during a 3-year period; the denominator is the number of people alive during those years. Mortality rates are generated by age and sex. The most recent triennium is 2009–11, with the calculated rates most representative of the year 2010. The rates for 2011–3 are therefore inferred, with a relative 1.4% compound reduction in mortality for each year since 2010 (as recommended by the ONS). The ONS rates are cross-sectional or ‘period’ mortalities, meaning that they represent rates at a given point in time, but cannot be used to generate an accurate survival curve from birth to death for an individual. To do so, one would need to use the period rates for each year throughout life. For instance, for someone born in 1981, one would use the infant rate for that year, followed by the 1982 rate for the next year of life and so on. Accurate ‘cohort’ survival curves can be calculated retrospectively, but future curves must infer rates.
  2. Different studies report different independent hazard ratios for common co-morbidities, such as myocardial infarction, stroke and heart failure. Methods to synthesise prognostic studies are nascent, with the Cochrane collaboration currently testing a model (for predicting postoperative nausea and vomiting). The calculator uses a hazard ratio of 1.5 for each of MI, stroke, peripheral arterial disease and heart failure, and 1.2 for angina and transient ischaemic attack when MI and stroke are, respectively, absent, whereas the ratio ranges between 1.3 and 2.5 in most studies.
  3. The glomerular filtration rate (GFR) falls with age, for which I have used the equation (0.003 × age2) − (1.017 × age) + 139.4. There are many equations that estimate eGFR from creatinine concentration and other variables. I have reported three eGFRs: MDRD; CKD-EPI; Cockcroft-Gault. The Cockcroft-Gault formula is of particular interest in the underweight and overweight, as it is the only one out of the three to adjust for body mass: for women ((140 − age) × mass × 1.04)/[creatinine]; for men((140 − age) × mass × 1.23)/[creatinine]. The mortality hazard associated with renal dysfunction increases by 1% for each ml 1.73 m−2 difference in eGFR between expected and calculated.
  4. The association of physical fitness with subsequent survival has been reported for a number of cohorts. A meta-analysis of these studies, despite a lack of methodology to support the synthesis, suggested that a shortfall in peak oxygen consumption, compared with that predicted (for someone that sex and age), was associated with a relative 15% increase in subsequent mortality hazard [32]. Predicted values for peak oxygen consumption (ml.kg−1) were adjusted for sex: women 3.5 × (14.7 − (0.13 × age); men 3.5 × (18.4 − (0.16 × age)) [30, 31]; method of exercise (0.85 for cycling compared to treadmill); and expected resting metabolic rate: women ((3.6 − (0.0367 × BMI) − (0.0038 × age) + (0.358))/2.72); men ((3.6 − (0.037 × BMI) − (0.004 × age) + (0.18))/2.61) [33].
  5. Various papers have suggested that peak oxygen consumption does not communicate all the prognostic information available from fitness: combinations of variables measured during cardiopulmonary exercise tests usually perform better than a single variable. Ventilatory equivalents for carbon dioxide (measured at the anaerobic threshold) have therefore been added to the calculation, predicted values of which vary with age and sex: women 22.09 + (age × 0.123), men 21.09 + (age × 0.123). Author's data indicate that relative mortality increases with the difference between observed and predicted CO2 ventilatory equivalent, by about 5% for each unit, independent of peak oxygen consumption.
  6. Second estimates of monthly mortality and associated survival curve are adjusted for body mass using the formula (0.0081 × BMI2) − (0.51 × BMI) + 8.54, which generates a nadir at a BMI of 32.
  7. The following equation estimates the annual risk (proportion) of an ‘× mm’ diameter AAA rupturing: (0.0001 × mm2) − (0.0068 × mm) + 0.1215, which for a 55 mm AAA is an annual rupture rate of 0.05 or 5%.