Cox proportional hazards model using different

Running the analyses using an incident cohort, censoring subjects at the date of their last appointment within the GPRD, or using death or liver transplant as the end-point led to no substantial differences in the adjusted hazard ratios for mortality either during the first year following diagnosis or subsequently (see Table 4-8).

Analysis using an incident cohort restricted the number of cases to 2787 and 26,176 appropriately matched controls. I showed a slightly higher hazard of death for subjects with compensated cirrhosis both during the first year

following diagnosis and subsequently compared to the initial analysis including both prevalent and incident cases. Patients with decompensated disease showed a slightly lower hazard of death in the first year following diagnosis but a greater hazard of death subsequently compared to the initial analysis.

When censoring subjects at either death, de-registration or date of last

appointment within the GPRD (as opposed to the end of the UTS period) there was a slightly lower hazard of death for both subjects with compensated and decompensated disease at both time periods studied compared to the initial analysis.

For the analysis using death and liver transplant as the outcome there was a slight increase in the adjusted hazard ratios for death in all subjects at both time periods studied compared with the initial analysis.

None of the hazard ratios observed in any of these three variations on analysis led to any significant differences to those calculated from the initial analysis.

Table 4-8 Cox proportional hazards model using different cohort definitions

Mortality rate (per 1000) Adjusted hazard ratio *

Control cohort Compensated cohort Decompensated cohort Compensated cohort Decompensated cohort During first year

Principal analysis 20.1 167.1 288.1 8.5 [7.6, 9.5] 14.2 [12.1, 16.6]

Incident cohort 24.1 212.3 309.7 9.0 [7.9, 10.2] 13.1 [11.0, 15.8]

Censoring at last GPRD appointment 22.5 171.1 293.8 7.9 [7.1, 8.9] 13.2 [11.3, 15.5] Death and liver transplant end point 20.1 172.8 300.1 8.8 [7.8, 9.8] 14.7 [12.6, 17.2]

Following first year

Principal analysis 19.0 73.8 133.0 4.3 [3.9, 4.6] 7.9 [6.9, 9.1]

Incident cohort 22.7 96.1 158.1 4.6 [1.4, 5.1] 8.8 [7.4, 10.5]

Censoring at last GPRD appointment 21.3 75.8 137.2 4.0 [3.7, 4.3] 7.4 [6.5, 8.5]

Death and liver transplant end point 19.1 80.6 142.0 4.6 [4.3, 5.0] 8.3 [7.2, 9.5] *adjusted for age and sex

4.4 Discussion

4.4.1 Key findings

Patients with a diagnosis of cirrhosis are at an increased risk of death than their age- and sex-matched general population controls with an adjusted hazard ratio for death of 5.8 (95%CI[5.5, 6.1]). Those with decompensated disease have an even worse prognosis with 5- year survival only 45.2% compared with 63.5% for those with compensated disease. The increased risk of death was particularly high during the first year following diagnosis but remained high subsequently, even following adjustment for some measure of comorbidity. The hazard ratio for death was highest in those with no reported comorbidity although absolute mortality rates were highest in those with reported comorbidities. Patients with a presumed aetiology of alcoholic cirrhosis had a worse survival at all stages than those with non-alcohol related cirrhosis.

4.4.2 Strengths and limitations

This study has been performed using a large, representative, population- based cohort of patients with cirrhosis alongside an appropriately matched general population based control cohort. These cohorts were identified relatively recently (1987 – 2002) and the results observed reflect the natural history of cirrhosis during this time period.

The size of the dataset used for this analysis has allowed the estimation of mortality rates and adjusted hazard ratios for death stratifying by severity of

Perhaps the most significant potential limitation of this analysis is related to the coding of the diagnosis of cirrhosis. As discussed in section 3.4.2 the inclusion of all patients with any mention of cirrhosis, oesophageal varices or portal hypertension within the medical records may have led to the inclusion of some patients who did not have cirrhosis. However, the same counter-

argument is applicable in this situation that it is more likely that there are cases with cirrhosis that are missing from the data source where the diagnosis of cirrhosis was unconfirmed and / or not fed back from secondary care to primary care records. This will potentially have led to an underestimate of the hazard ratio for death as cases with undiagnosed cirrhosis may be included within the control population.

In addition to the potential for misclassification of the diagnosis of cirrhosis it is possible that there exists further misclassification in the accuracy of the

severity of cirrhosis as compensated or decompensated. Whilst the results of the validation described in section 3.3.4 lend credence to the diagnosis of cirrhosis itself, with evidence that most diagnoses occurred in secondary care, the prohibitive cost of validation within the GPRD system did not allow for the validation of the additional signs and symptoms of ascites and gastrointestinal bleeding used to define decompensated disease. Indeed, these clinical symptoms may only be recorded if they are of obvious clinical relevance to the GP. Ascites and gastrointestinal bleeding are however serious complications of cirrhosis and as such one might expect that these events would be

considered important and therefore recorded within the primary care record, either following hospital correspondence or from an individual patient

consultation. It is possible that the recording of ascites, if it occurred in the absence of further information from hospital, would only be large-volume, clinically significant ascites visible at the general practice appointment rather

most likely to act through misclassifying patients with decompensated cirrhosis as having compensated disease this will have led to an overestimate of the hazard ratios for death for people with compensated disease and a

corresponding underestimate of the hazard ratios for death for people with decompensated disease.

The size of the database used has allowed the description of the mortality experience of patients with both alcoholic and non-alcohol related cirrhosis alongside their matched controls. The small numbers of cases with other presumed aetiologies meant that it was not possible to model the hazard of death for other causes of cirrhosis with any precision.

It could be argued that the inclusion of both incident and prevalent cases in the cirrhosis cohort might lead to the introduction of survival bias as those cases who are prevalent have, by definition, already survived a particular length of time to still be included in the analysis. As such, I ran the principal analysis using only those subjects who were considered incident as defined in section 3.2.2.1. The estimates of the mortality rates and therefore the hazard ratios for death were slightly higher than those seen when including all

incident and prevalent cases, apart from subjects with decompensated disease in the first year following diagnosis who showed a slightly lower hazard ratio for death. However, all estimates remained within the 95% confidence interval of the principal analysis. The results I have reported including all incident and prevalent cases probably more accurately reflects the real world of clinical practice within the general population and allows for the communication of results that are directly valid to the patients an individual

Though the data available have allowed for an appropriate individually- matched adjustment for some confounders (age and sex) it was not possible to examine the potential associations or modifying effects of other variables either because they are not available in the GPRD at all e.g. any measure of socio-economic status which is known to be associated with mortality, or because there was such a high proportion of missing data. Recent studies have suggested an interaction between raised BMI and alcohol consumption leading to an increased risk of death from liver disease.91 Additionally it was not possible within this dataset to examine the cause of death of these patients as this information was not systematically available.

Of note is the discrepancy in the proportion of patients who were recorded to be alcoholics before the diagnosis of cirrhosis (30.4%) and the proportion of patients assigned a presumed aetiology of alcoholism (50.9%). Aetiologies were ascribed using information contained in the whole medical record (including after the diagnosis of cirrhosis). This would therefore suggest that roughly two-fifths of those patients with alcoholic cirrhosis were not known to be (or at least not recorded as) alcoholics before the onset of their disease. However, when taken in context to the 96.0% of controls who did not have their alcohol status recorded before the date of pseudo diagnosis, it is perhaps encouraging that so many patients with cirrhosis had already been identified as alcoholics.

4.4.3 Comparison with previously published work

In the UK the most recent published study on cirrhosis mortality observed the mortality of patients with all chronic liver disease who had been admitted to hospital in a small geographically distinct region of the UK between 1968 and 1999.41 This study reported an SMR for one year follow-up of 16.3. This figure is higher than the 1-year adjusted hazard ratios that I have reported for either patients with compensated or decompensated disease.

Other large cohort studies from European centres have similarly shown worse survival than I have reported in this chapter. In Denmark between 1995 and 2006 a hospital cohort of 14,976 patients with cirrhosis described a 1-year survival of only 65.5% and 5-year survival of 37.5%, notably lower than that I have reported.92 This study was not able to categorise patients into

compensated or decompensated disease and additionally did not report the hazard ratios for mortality compared with their matched control cohort.

Previously the same authors had reported SMRs for patients hospitalized with cirrhosis compared with the general population of 32.3 for the first year

following diagnosis and 5.7 thereafter.46

These poorer survival estimates are perhaps unsurprising as these previous studies used cohorts of patients admitted to hospital. Whilst the diagnosis of cirrhosis in my study population is likely to have come from a hospital

diagnosis it does not necessarily follow that these patients were hospitalized for this diagnosis to occur. As my study is a population-based cohort the results seen are also unlikely to have been affected by the variation in referrals and follow-ups seen in cohorts selected from secondary care. As such it is probable that my results can be more widely generalised to patients diagnosed with cirrhosis including those ambulatory patients who have not

Perhaps the most commonly referenced figures in the literature looking at the mortality experience of patients with cirrhosis based on the severity of disease are those based on 1155 consecutive patients admitted to a single hospital in Sicily during the 1970s and 1980s which reported 6-year survival of 54% and 21% in patients with compensated and decompensated cirrhosis

respectively.44 It is possible that the survival figures I report are better than those reported in this smaller study, particularly for those patients with

decompensated disease, because there have been significant improvements in the management and outcomes of the complications of cirrhosis in the intervening decades. This has been shown for oesophageal varices in studies in both the USA and more recently Sweden.5860 Additionally, the proportion of cases ascribed as alcoholic cirrhosis is lower in the Sicilian population than in my study population with a correspondingly higher proportion of patients with hepatitis B virus. It is therefore possible that the survival figures reflect a different spectrum of aetiology of cirrhosis in the two populations.

Other estimates of mortality in patients with cirrhosis in the UK come from much older studies. Saunders et al report the 5-year survival of patients with alcoholic cirrhosis, 36%, and non-alcohol related cirrhosis, 14%.37 An earlier study, also from Birmingham, covering patients with cirrhosis in a single hospital between 1959-64 reports similarly low 5-year survival figures of 14.4% after first admission to hospital, again with patients with alcoholic cirrhosis having a better prognosis (5-year survival of 20% compared with 5.1% for patients with cryptogenic cirrhosis).43

The survival figures I have reported are better than these two previous studies but conversely show that survival with non-alcohol related cirrhosis confers a

4.4.4 Conclusions

In this study I have shown that the survival of patients with a diagnosis of cirrhosis, whilst better than previously reported studies,37414344454647485793 is still poor with 1-year survival of 83% and 5-year survival of 60%. This study represents the most contemporary estimate of the mortality experience of patients with cirrhosis and also allows comparison with the general population. Even in those patients with no comorbidity the adjusted hazard ratio for death is significantly higher than that of the general population.

I have been able to describe the mortality experience of patients with

compensated and decompensated disease, showing a substantial increase in mortality for those with decompensated disease at diagnosis.

Defining decompensated cirrhosis as the presence of ascites or GI bleeding appears to accurately reflect an increased probability of death compared to compensated cirrhosis. Although there is the possibility of misclassification of patients with decompensated disease as compensated as differences in follow-up at individual sites may exist, these general population based data provide an „average‟ estimate of the mortality experienced by patients with cirrhosis. These figures will be of particular importance for service providers and for individual clinicians and general practitioners to communicate to their patients.

For the purposes of service provision and planning it is important to note that even patients with compensated disease had an increased hazard of death compared to the general population. This is in contrast to the assumptions made for the purposes of the cost-effectiveness analysis carried out in a

that there is no excess mortality in those patients with compensated disease compared with the general population.49

An aetiology of alcoholic cirrhosis remains a particularly bad prognostic indicator with mortality in patients with alcoholic cirrhosis nearly double that of patients with non-alcohol related cirrhosis. With roughly 60% of all patients with alcoholic cirrhosis being known to their GPs as alcoholics before the date of diagnosis of cirrhosis this may provide an opportunity for intervention(s) leading to behaviour and risk modification.