Background: Myasthenia gravis (MG) is a rare neurological disorder, which can be life threatening. Although death is a rare outcome, evidence does not exist as to whether neurological care leads to any better outcome than care by other specialties.
Methods: A matched nested case control study sampled from all public sector hospital admissions in England with a primary diagnosis of MG from 1991 to 1999. Cases were defined as MG admissions which resulted in death and controls were other MG admissions, matched on sex, age (±2 years) and date of admission (±20 days) that were non-fatal. From a total of 18 251 finished consultant episodes with a mention of MG, we were able to create 196 matched sets with 196 fatal admissions and 788 control admissions.
Results: Admission under a neurologist was associated with a 69% reduced risk of death (OR 0.31, 95% CI 0.22 to 0.44; p<0.001). This was only slightly attenuated after adjustment for a variety of patient related and hospital covariates (OR 0.37, 95% CI 0.23 to 0.62; p<0.001).
Conclusions: This is the first evidence that patients admitted with MG are far less likely to die if they are under the care of a neurologist. We cannot determine whether this is because of better management per se or because neurologists are usually based in specialist centres and may have better intensive care support, or both. Alternatively, this may be a result of “selection bias” so that neurologists select less seriously ill patients.
Statistics from Altmetric.com
Myasthenia gravis (MG) is a relatively uncommon disease caused by impaired neuromuscular transmission and is characterised by fatiguable muscle weakness. There is a wide spectrum of disease severity, ranging from purely ocular disease to a myasthenic crisis, where weakness of the bulbar and respiratory muscles can rapidly lead to respiratory failure and death. The introduction of immunosuppressant treatment and intensive care support of people in a myasthenic crisis in the 1970s and 1980s had a significant impact on disease mortality, with a halving of mortality from 15% to 7%1 and an increase in disease prevalence.2 More recent data, however, suggest that the crude mortality rate has been increasing although case fatality appears to have remained fairly constant.3 One explanation for this could be the greater recognition that MG is not just a disease of young women but, if anything, is more common in the elderly4 5 who may have been previously undiagnosed6 and have worse survival.3
In this world of evidence based medicine, empirical scepticism means that it is justifiable to question the effectiveness of specialist care compared with that provided by a generalist. Few studies have even attempted to evaluate the usefulness of neurological care. A review of neurological inpatient consultations noted modest benefit in terms of generating a new differential diagnosis or ordering a new procedure that ultimately proved useful in making the diagnosis.7 Reducing diagnostic uncertainty may be helpful to patients and allow better prediction of prognosis but may not improve patient outcome or quality of life. In a similar retrospective review in the UK, approximately 30% of neurological inpatient consultations led to a change in management.8 This is more encouraging but no patient outcomes were reported. Given the rarity of myasthenia gravis, most clinicians, including some neurologists, are likely to have little experience in managing such patients. It therefore seems reasonable to suppose that inpatient care by a neurologist may improve mortality risk, especially during a myasthenic crisis. However, undertaking a randomised controlled trial of neurologist versus generalist care would be impossible given the very large sample size requirements. It is also unlikely that patients would be happy to be randomised, given the option, as a lay perspective is likely to favour specialist care and therefore it would probably be regarded as unethical. One is therefore forced to examine observational data to test the hypothesis that specialist care may reduce the risk of death.9 We have undertaken a nested matched case control study within a large administrative database to examine whether mortality risk for MG is reduced by specialist care.
Hospital episode statistics data
The Hospital Episode Statistics (HES) database is a record of every hospital admission in England, excluding the private sector (see the following web link for more detailed information http://www.hesonline.nhs.uk/Ease/servlet/ContentServer?siteID = 1937&categoryID = 537). An anonymised extract from the HES database was provided which identified cases with MG using ICD-9 code 358 (1991–1994) and G70 in ICD10 (1995–1999) for the financial years 1990/1991 to 1998/1999. Each patient can have one or more admissions. Within an admission, there may be one or more finished consultant episodes (FCEs). An FCE is when a patient is transferred to another doctor within the same admission. In other words, if a patient was admitted under an orthopaedic surgeon for a hip replacement, but then had a heart attack and their care was transferred to a cardiologist prior to discharge, this would appear as two FCEs. This database contains no personalised information or unique identifier, but it is possible to create a pseudo-unique identifier by using a combination of sex, date of birth (encrypted) and postcode (encrypted) within a disease category. In this way, the same patient may be identified with multiple FCEs within the same admission as well as repeat admissions on different occasions.
Definition of cases and controls
Cases were identified by using the code for discharge method, which includes death as a possible outcome. However, HES does not record the cause of death. Clearly, patients with MG are also admitted for other disease such as a myocardial infarction, which may result in death unrelated to their MG. Therefore, we only included records where the diagnosis in the primary diagnostic field was MG or the operative code was B18 (excision of thymus gland). As autoimmune myasthenia is uncommon before puberty; all subjects under the age of 10 years on admission were removed to minimise confusion with congenital myasthenic syndromes.
We selected controls from other MG admissions around the same time as the cases. This is technically known as “incidence density sampling” and means that the calculated odds ratios (OR) are equivalent to the rate ratios.10 Controls were generated by comparing cases with the complete dataset and matching them on the following criteria: (a) sex, (b) the control date of admission had to be within ±20 days of the case admission and (c) the control had to be within ±2 years of the case age. For each case, we selected all controls that matched on the above criteria to enhance our statistical power. Some cases had more than one FCE and therefore also appeared in the potential control list during the same admission that resulted in their death; these subjects were removed. In addition, genuine controls could appear more than once as they had experienced more than one FCE in an admission; these duplicate control records were also removed. Finally, it was not always possible to find a control that matched all the criteria for a case. Where this occurred, the case was dropped from the final dataset as they could not contribute to the analysis.
Explanatory and confounding variables
We hypothesised that both patient (clinical and sociodemographic) and provider related (clinical specialty, hospital experience) variables may influence the risk of death. Both age and sex were matching factors, so further adjustment for these potential confounders was unnecessary. Our knowledge of the patients’ clinical state was limited, but we constructed three “clinical” variables that might influence the likelihood of mortality. Firstly, admissions were classified as either urgent (emergency admission or emergency transfer) or routine (elective admission or elective transfer), indicating whether the patient was unwell at the time of admission. We were also able to identify the number of repeat admissions that a patient had experienced over the 9 years’ worth of data from the complete data extract. We categorised patients’ history of admissions into three groups (0 or 1, 2, 3 or more repeat admissions). We assumed that patients with recurrent admissions had less well controlled myasthenia than those with fewer admissions, although some of these admissions may have been for non-MG morbidity. Finally, a patient’s comorbidity was determined by calculating the Charlson Index11 for each admission using all of the diagnostic data from the multiple discharge codes. The Charlson Index creates a weighted score based on the presence or absence of certain diseases known to increase mortality. This was categorised into three groups based on the total score (0, 1–2, 3 or more). We used the Townsend Score of area deprivation12 for each individual’s ward of residence as an area based proxy measure for individual socioeconomic status. This was then grouped into quartiles (quartile 1 most affluent, quartile 4 most deprived).
We derived several provider related variables that might indicate quality of care. Firstly, we classified whether patients were under the care of a neurologist or another specialty, this being our main exposure variable. Secondly, we categorised the experience of hospitals in caring for MG patients. Using the complete dataset, we counted the total number of MG admissions, removing duplicate records for a single admission episode, by hospital code. From this, we generated an “experience” variable with three levels (<50, 50–99, 100+ admissions). We decided to take into account crude geographical differences in care by classifying the region of treatment into North and South England.
For the basic descriptive data, we have presented percentages and means by case control status ignoring the matched sets. We used multivariable conditional logistic regression analysis to calculate OR (95% confidence interval (CI) and p value) for mortality by our key explanatory or confounding variables. An OR >1 indicates an increased risk while an OR <1 indicates a decreased risk associated with that specific variable. As sex, age and admission date were matched in the case control sets, these were not included in our models. We did not specify any interactions a priori. However, to examine for such interactions, we initially ran stratified analyses to see if there was any heterogeneity in the OR by any potential effect modifier. We only undertook a formal test of interaction if these results suggested a possible interaction. We did not use any Bonferroni adjustment for multiple testing as our primary exposure variable (specialist care versus other) was specified a priori and we were only interested in seeing if the effect estimates altered after adjustment for a variety of confounding variables.
Basic demographics of sample
Between 1991 and 1999, there were 18 251 finished consultant episodes with a valid discharge code. Mean age of the patients was 59.1 years (median 64) and there was an excess of women compared with men (53.9 vs 46.1%). In this original sample there were 770 admissions that resulted in death (4.2%). By limiting the sample to only subjects with a primary diagnosis of MG or thymectomy, there were 10 087 records, which included duplicate patients with more than one FCE within an admission and 224 deaths (2.2%). Thirty-seven records had no specialty coded, leaving 10 050 records. A total of 6751 episodes (67.2%) were under the care of a neurologist, 1825 episodes (18.2%) under a general physician or geriatrician, 498 episodes (5.0%) under a cardiothoracic surgeon and 976 (9.7%) under another specialty code. After matching with control subjects, there were 196 cases (28 cases dropped as no suitable control) and 788 controls (984 observations). The number of controls per case varied from 1 to 12 with a mean of 5.4 controls (median 5 controls).
The basic descriptive data for cases and controls are given in table 1. Although the number of men and women appears to be imbalanced, this is not because the matching process was unsuccessful. Male cases generated more male controls than female cases, resulting in a higher percentage of men in the control than the case arm. As expected, patients admitted as emergencies and with more comorbidity had a much higher risk of mortality. There was no obvious trend, however, with a past history of admissions or with socioeconomic status, as measured by the Townsend Area Deprivation Score. Increasing hospital experience with MG was also associated with a marked linear reduction in risk for mortality (36%) per change in grouping. There was no marked difference in risk by area of treatment.
The simple OR for specialist care showed a marked reduction in mortality (69%) that was unlikely to be due to chance (p<0.001) (model 1). We explored whether the effects of neurologist care persisted after fitting two further models; model 2 included patient related variables and model 3 included patients and provider related variables (see table 2). These are clearly potential confounding factors as patient and hospital related factors may both explain risk of mortality as well as likelihood of being under the care of a neurologist. These adjustments attenuated the benefits of neurological care by only a modest degree. The effects of comorbidity remained strong while the influence of type of admission was weakened and hospital experience was no longer predictive.
We wondered whether the apparent “protective” effect of being under the care of a neurologist was artefactual because of day case admissions for undertaking an edrophonium (Tensilon) challenge test to confirm the diagnosis of new cases. Such admissions would be likely to be under a neurologist and are unlikely to have resulted in a fatal event. The results, however, were essentially unchanged after excluding any cases and controls with length of stay of <1 day (OR 0.32, 95% CI 0.22 to 0.45; p<0.001). The protective effects of neurological care were seen when stratified by elective and emergency admissions but were less marked for emergency admissions (OR for neurological care and elective admission 0.20, 95% CI 0.10 to 0.40 (p<0.001); OR for neurological care and emergency admission 0.52, 95% CI 0.31 to 0.85 (p = 0.01)). A formal test for interaction showed some evidence against the null hypothesis that there was no interaction by type of admission and neurological care (p = 0.02). Other stratified analyses, by comorbidity, past admission history, area of treatment, hospital experience and age group of patients made little difference to the protective effect of neurological care. We also repeated the analysis excluding cases or controls who were under either a general surgeon or cardiothoracic surgeon, as surgical admissions may have a greater risk of mortality. However, this did not greatly alter the results (OR 0.29, 95% CI 0.21 to 0.42; p<0.001). To examine the robustness of the association, we dropped all observations where there was any other comorbidity or operative code in any of the diagnostic or operative fields, except MG in the primary field. This reduced the data set to 334 subjects. However, many of the matched case control sets either had no case event or there was a case but no control, leaving us with 68 observations from 20 sets. The OR for being under a neurologist remained highly protective (OR 0.12, 95% CI 0.02 to 0.59; p = 0.009). One possible scenario, which could produce an artefactual protective effect for neurological care, is if patients switch from neurological care to a general physician for the last FCE. We therefore repeated the analysis using only the last FCE for controls, by regenerating a new set of matched case control sets. By definition, the cases who died cannot have any further FCEs. The crude re-analysis found almost exactly the same results (OR 0.32, 95% CI 0.22 to 0.45; p<0.001).
This is the first study to report whether patient and provider related risk factors alter the risk of inpatient mortality for patients with MG. It specifically provides empirical evidence as to whether neurological care makes any difference on a hard endpoint such as mortality. Mortality rates for MG are small, given the rarity of this disorder and its case fatality. Only 1.2% of all FCEs resulted in death. This makes it almost impossible to study the determinants of mortality with either a randomised controlled trial or cohort design. Even a conventional case control study would require multicentre national or international collaborations to recruit enough patients with MG who died. The use of a nested case control study within a large routine dataset provides an opportunity to test a hypothesis that would be difficult if not impossible to evaluate using other designs.
The effect of being under the care of a neurologist was marked, with a crude reduction in risk of 69%. If such an analogous benefit was seen with a new drug therapy, it would be hailed as a remarkable new treatment. In addition, increasing hospital experience with patients with MG was associated with a reduction in risk. In our multivariable models, this no longer predicted mortality risk, possibly as the benefit seen in experienced hospitals was mediated through the care of a neurologist. Because the HES database does not currently link deaths to death certification records, we cannot be sure that all deaths in the case control study were due to MG, although it is likely that MG was an indirect contributor. For example, the case fatality rate from heart attack or stroke may be greater if the patient’s myasthenia is poorly controlled. Not surprisingly, emergency admissions and comorbidity were strong predictors of mortality, although area deprivation and past admissions did not predict death.
There are several possible explanations as to why patients who are under a neurologist appear to have a lower mortality risk. This may have been a chance finding although the small p value provides strong evidence against the null hypothesis. As this it the first time such a finding has been reported, it is important that this observation is replicated in other datasets. There may be some bias in ascertainment. However, our data were collected by a routine administration system. It is unlikely that admissions that did not result in death would be any more likely to be registered as under the care of a neurologist compared with another specialist. Despite trying to only select admissions for myasthenia using the primary diagnosis field, it is possible that the primary reason for admission and hence death was not MG but another comorbidity such as myocardial infarction, which would be usually cared for by general physicians and not a neurologist. This would result in an apparently artefactual benefit due to neurological care. However, even when we limited our analysis to non-surgical patients and only cases with no other diagnostic or operative code, we still found a beneficial effect of neurological care. In addition, the total number of fatal events in our dataset (224 deaths) is compatible with that recorded by the Office of National Statistics for the same time period with an underlying cause of death of MG (310 deaths). Another potential explanation is that our results reflect selection bias. If patients in a life threatening state are reviewed by a neurologist but are not transferred to their care, this would then be coded as death under another specialty. This would be especially true in a hospital with no on-site neurological care where a neurological opinion may be sought by telephone or a ward referral. Similarly, patients who are under a neurologist at a specialist centre but unlikely to recover may be transferred to a generalist for terminal care. Thus we cannot exclude the possibility that neurologists select which patients to look after assuming they are not admitted directly under their care or that some patients die before neurological assessment so that the neurologist cannot formally take over responsibility for their care. We adjusted for both comorbidity and whether the admission was an emergency in our analyses, although the use of diagnostic information from routine datasets is less reliable than that from more detailed hospital chart reviews.13 There remains the likelihood of residual confounding, particularly with regard to other hospital related factors. For example, it is possible that neurologist are based at hospitals with better intensive care units and the apparent benefit of neurological care is actually a marker of this facility. The results may have occurred because of some management artefact. For example, we considered that patients admitted for a routine edrophonium (Tensilon) challenge test, and therefore unlikely to die, would be differentially admitted under the care of a neurologist and hence be more likely to be represented in the control group. However, excluding day cases made no difference to our results. We did find an interaction between neurological care and type of admission, but there still remained a beneficial effect even for just emergency admissions. One might argue that neurological care should have been more important for emergency admissions, but it is possible that patients with MG not admitted in crisis under other clinicians deteriorate because of suboptimal management of their medications either around admission or after some complication has developed.
Finally, it is possible that neurological care is truly associated with a lesser risk of mortality as neurologists will be more experienced in dealing with myasthenic patients and may be more proactive and aggressive in their management. There is evidence that this is true for other non-neurological diseases such as cystic fibrosis,14 15 and rare cancers have a better prognosis when managed by an experienced clinician and/or unit that treats a greater volume of cases.16–18
While mortality from MG is rare, our results, if valid, have important implications for health care delivery—all patients with MG admitted should have joint care from the admitting clinician and a neurologist experienced with MG, especially if they are compromised by their MG. Our data cannot determine what proportion of deaths, if any, would have been avoidable if better care had been provided. To be sure that the results are not artefactual it is important that they are replicated, with data relating to the underlying cause of death but also extended by undertaking some form of confidential enquiry, as seen with perioperative surgical mortality.19 This would enable us to determine what proportion of deaths are potentially avoidable by providing more specialist care.
The Hospital Episodes Statistics (HES) data were made available by the Department of Health. We wish to thank Davidson Ho for extracting the HES data and Professors George Davey Smith and Shah Ebrahim for helpful comments on a draft manuscript. HES analyses conducted within the Department of Social Medicine are supported by the South West Public Health Observatory.
Competing interests: None.
Ethics approval: We did not have formal ethical approval as this work involved the secondary data analysis of routinely collected anonymised data. The Department of Health reviewed and approved our departmental request to access the data for research purposes.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.