Explanation of statistics in “Clinical outcomes” paper

Recently I had posted about this O’Sullivan “Clinical outcomes” paper.  And I had asked if anyone knew statistics and could understand what “older age of onset” means (the abstract indicates in MSA “older age of onset” is a factor “predicting shorter disease duration until death”) and what “early autonomic dysfunction” means (in MSA “early autonomic dysfunction” is a factor predicting shorter survival).

Well….local support group member Ted is a biostatistician!  He has answered these two questions in addition to providing a useful (and short) summary and some interesting comments about what is statistically significant.  Here’s Ted’s email below.  It’s just in time for the support group meeting tomorrow:  I know some of you MSA caregivers are really interested in this article!  Thanks Ted!



From Ted, local Brain Support Network group member

This is a complicated paper, so I’ll try just to summarize the parts I think are interesting, without adding very many thoughts of my own.

First, the very short version: MSA tends to start at a younger age than PSP.  Survival after getting the disease averages about 8 years for both diseases.

If you have PSP, you will tend to live longer if your early symptoms are more like typical Parkinson’s, if you are younger when you get the disease (though the difference isn’t large), if you are female, or if it is a comparatively long time before you hit your first “clinical milestone”, a list of bad things that happen as the disease progresses (I have a list below).

If you have MSA, you will tend to live longer if you are admitted to residential care at some point, if you do not have early autonomic dysfunction (various symptoms related to urination, digestion, sweating, blood pressure, and sexual function, within the first two years after onset; list below), if you are younger when you get the disease (again, not a big difference), if you are male (but this is a little complicated, as I explain below), and if it is a comparatively long time before you reach your first clinical milestone.

For both diseases, the degree of response to L-dopa does not appear to be associated with survival, at least in this study.

In this paper, the diagnosis of PSP or MSA is based on autopsy results.  The authors note that the accuracy of diagnosis during the patients’ lifetimes was less than 100%, and that PSP and MSA are often hard to distinguish during the patient’s lifetime.   All the conclusions presented were based on a set of 110 PSP patients and 83 MSA patients who had sought treatment and permitted autopsies at the Queen’s Square Brain Bank for Neurological Disorders.  This group of patients may not be exactly comparable to our own group of family members with PSP and MSA.

Now the less short version, in answer to your two questions:

Older age of onset:  First, “age of onset” means the age at which the first symptom attributable to PSP or MSA appeared in the patient’s medical record.  This is usually earlier than age at diagnosis (by 3-4 years on average; table 1).  “Older age of onset” does not refer to any specific cutoff age, but instead to the authors’ finding that patients who are older at onset tend not to live as long as patients who are younger at onset.  The “hazard ratio” for age of onset increases at 5% per year for both MSA and PSP (table 2).  The “hazard” is the rate at which patients die, so it would be measured in units like 10 deaths per year per 100 patients.   So, for example, patients with age of onset 64 would have a hazard approximately 20% higher (hazard ratio 1.2) than that of otherwise comparable patients with an age of onset of 60 (4 years younger at onset times 5% per year).

This is a small effect compared to some of the others shown in the table.  As an example, for PSP patients, male gender has a hazard ratio of 1.7 (meaning men have a 70% higher hazard than comparable women), which is equivalent to about an 11-year age difference at onset (the actual calculation is somewhat more complicated than just multiplying number of years times 5%).  So for PSP patients, a male with onset at 60 would have about the same hazard as an otherwise similar female with onset at 71.  Age isn’t in the same league as the big hazard ratios in the table, such as early autonomic dysfunction for MSA patients (hazard ratio 6.0).   I don’t think it’s clear from the paper how much of the age effect is due to differences in the progression of the disease versus just the increased hazard you would see in any comparison of older people versus younger.

Early autonomic dysfunction:  The autonomic nervous system regulates bodily functions like heart rate, respiration, digestion, salivation, urination, and sexual arousal.   For the purposes of the paper, autonomic dysfunction was defined to mean either an abnormal autonomic function test result, or (and I think this was what was used for the vast majority of cases) having two or more of the following set of symptoms, as reported by the patient:

  1. Frequent or urgent need to urinate, or “nocturia without hesitancy”, which sounds like it might mean either bedwetting or urgently needing to get up during the night to urinate; I’m not sure of the exact medical usage;
  2. Chronic constipation;
  3. Postural hypotension, presenting either as a complaint from the patient or just from observing a sufficient difference between sitting and standing blood pressure in the doctor’s office;
  4. Sweating abnormalities; and
  5. Erectile dysfunction.

Autonomic dysfunction was classified as early if it appeared within 2 years of onset (as defined above, first attributable symptom in the medical record).  The presence of erectile dysfunction on the list is a complicating factor, because it is a symptom that only men can experience, which means it is easier for a male patient to be classified as having early autonomic dysfunction than a female patient.  The authors mention this, and note that, as one would expect, a higher proportion of male patients had early autonomic dysfunction.  I’ll say a bit more about this later.

Finally, some unsolicited bonus commentary:

If you read the paper yourself, the word “significant”, or “statistically significant”  is used in a technical sense different from the everyday meaning.  A statistically significant effect is one for which the evidence is strong.  The effect itself may be rather weak (or it may not be).  So, for example, in table 2 the effect of age of onset is statistically significant, but it doesn’t have a very large effect on the hazard ratio compared with, say, the RS phenotype for PSP.

OK, then going over table 2, the biggest effect for PSP seems to be the RS phenotype.  This means that time to survival is shorter for patients whose main symptoms during the first two years are falls, cognitive dysfunction, supranuclear gaze palsy, abnormal saccadic (quick) eye movements, and postural instability.  PSP-P patients, whose first two years are more along the lines of typical Parkinson’s, with slow movement, tremor, some response to L-dopa, asymmetric onset and limb stiffness, tend to survive longer.  The hazard ratio is 2.37.   There were some patients for whom it was not clear which group they fell into, and they were excluded from the analysis.  Male gender is associated with shorter survival (hazard ratio 1.7), as is older age of onset, but for age it’s not a big difference.  Finally, patients who reach their first clinical milestone later tend to survive longer.  This is another per-year effect, with each year of delay for the first milestone reducing hazard by 20%.  On average, PSP patients’ time to first milestone varies by two or three years (bottom rows of table 3, showing a standard deviation of 2.7 years for PSP), so at 20% per year this is a pretty big effect.

For MSA, the big effects, which are really big, are early autonomic dysfunction (hazard ratio 6.0), and female gender (hazard ratio 3.0).  Both of these shorten survival, and I think they have to be viewed as a unit because of the erectile dysfunction symptom, which makes it easier for men to be classified as having early autonomic dysfunction.  It seems to me the high estimated hazard ratio for the female group may partly be explained as a “penalty” the female group pays for the greater proportion of undiagnosed early autonomic dysfunction in that group (this is my opinion, not the authors’, and I could be wrong).  Again for MSA patients, age of onset and interval to first milestone show up, with similar results.  Finally, not being admitted to residential care has an estimated hazard ratio of 2.8, so (perhaps surprisingly) residential care seems to increase survival for MSA.

The clinical milestones are a list of seven bad things that may happen to patients between disease onset and death: (1) frequent falls (2/year or more); (2) dependency on a wheelchair; (3) unintelligible speech; (4) severe dysphagia (difficulty swallowing); (5) use of a urinary catheter; (6) cognitive impairment; and (7) entering residential care.  Most patients experience at least a few of these (figure 2), but very few experience all seven.  Table 3 is a big summary of roughly when each milestone tends to occur (among patients that experience it), and a comparison of frequency of milestones between the disease groups.  Figure 4 plots the average times at which milestones occur, relative to the disease course, for Parkinson’s, PSP, and MSA.  The figure also shows time of diagnosis.  I’m a little dismayed to see how MSA patients seem to have a pileup of milestones at the end of life (although again, most patients don’t experience all milestones, and the times shown are the average for patients that do experience it).   I like the way the figure displays the average age of onset and duration for the three diseases.

Here’s the figure four of the O’Sullivan paper:

O’Sullivan et al, Fig. 4

Well, I think that’s all I have to say.  I hope it’s not too incomprehensible.  Please feel free to distribute this, or parts of it, to the group.