OK, I’m going to stipulate right off the bat that I am allowed at most three jokes related to Canadian stereotypes throughout this piece. Gotta ration them out, wouldn’t want to go overboard like a Canadian at Tim Horton’s. OK that’s one. I’d apologize for it, but I don’t want to blow my second Canadian joke right away (that’s two). But right — medicine.
Care of patients with cystic fibrosis (CF) has been revolutionized over the past 40 years, from basic interventions like aggressive early childhood nutrition with a high fat diet to the advent and spread of lung transplantation. Median expected survival has climbed consistently over this period. A driving force of these innovations in CF care has been direct comparisons of outcomes between centers, with observed differences leading to quality improvement initiatives and standardization of care.
So when Anne Stephenson and colleagues examined differences in median survival between the US and Canada on the basis of these nations’ respective CF registries, it was less revolutionary on the basis of aim, but still very much so on the basis of methodology and the centers included. What they found can inform our national debate around health care policy as well as directions for future research in cystic fibrosis care.
If You’re Only Going to Read One Paragraph
In a comparison of two national registries of patients with cystic fibrosis, Canadian patients had a statistically greater median survival than their US counterparts (50.9 years vs 40.6). When adjusted for baseline patient factors and disease severity, the relative risk for death was 0.66 (95% CI 0.54 to 0.81) in Canada versus the US. While this may be partially explained by differing rates of transplantation (10.3% in Canada vs 6.5% in the US) and differing post-transplant survival, this is unlikely to explain the magnitude of the effect. When the same US patients were stratified by insurance status, Canadians did not have a statistically greater median survival than patients with private insurance, but did have lower risk of death compared to patients on Medicare/Medicaid, and had an even lower relative risk in comparison to uninsured patients (HR for all Canadians vs uninsured US patients 0.23, 95% CI 0.14 to 0.37). This is a great example of how access to health care has real-world implications for peoples’ lives, especially those with preexisting conditions, and can actually impact length of life. This example should inform debate around health policy in the US.
A Bit of Context
Between 2000 and 2010, mortality for US patients with CF fell by 1.8% per year. Despite this consistent improvement in life expectancy, it’s well known that some centers do better than others. The practice of comparing outcomes between centers has led to standardization of care for CF patients. A recent comparison of PFTs among children and young adults in the United Kingdom versus the US, for example, revealed that US children had statistically better lung function. This was associated with differing rates of inhaled mucolytic prescription, and is a jumping off point for expanding this intervention across the UK.
The reported median age of survival in the US database was 36.8 in 2011, while the Canadian registry reported a median of 48.5 years. However, as we’ll talk about below, these reported numbers may not be good candidates for direct comparison. No prior study has delved into the registries themselves to verify the magnitude of this discrepancy, or to examine factors which may underpin its existence.
If this is a professional interest for you, the editorial accompanying this piece in Annals and its various citations is a great place to start learning more.
The US Cystic Fibrosis Foundation Patient Registry (CFFPR, N = 32,699) and Canadian Cystic Fibrosis Registry (CCFR, N = 4,662) each collect data from dedicated cystic fibrosis centers in their respective countries. The data collected is extensive, and includes age at diagnosis, treatments prescribed, genotype, relevant microbiologic data, PFTs, extrapulmonary manifestations of disease, and, most importantly for our current study, age at death. Why didn’t I shorten that list and just tell you the most important pieces of data they collect? Guys: that is the shortened list. Seriously there is a lot of information in these registries. The authors reviewed and standardized this data collected from 1986-2013 for purposes of comparing median survival between the two countries.
Registry data has its limitations, especially when comparing data collected by two different groups. The authors of our current study have previously described how different methods of accounting for missing data, dealing with patients lost to follow up, and following outcomes after lung transplant can dramatically change the reported median survival. So they are careful to dive into the source data directly rather than rely on any reporting from the registries, using patient-level data to perform an apples-to-apples comparison of the two populations.
One interesting statistical point — if you look at Table 1, a comparison of demographic and clinical factors between the two registries, you will notice that there are numerous factors noted to have a P value < 0.001 in comparison between the US and Canada. The authors correctly note that with sample sizes this large, very small differences may nonetheless be statistically significant (see for example the 47% Canadian vs 43.6% US delta F508 homozygous population). To avoid over-reporting of differences that may not be clinically relevant, the investigators instead elected to use a standardized difference measurement > 10 as their threshold for significant difference between countries. Standardized difference is the ratio of the difference in means to the difference in standard deviations between the two groups. Without getting bogged down in statistics, suffice to say that this is a measure of effect size rather than simply the presence of a statistical difference (the actual difference between the means, not just that their confidence intervals do not overlap, dictates the threshold for inclusion). Thus, only factors that differ significantly in magnitude between two groups will be picked up by this method. This is a nonstandard use of standardized difference (see what I did there?), but does have the advantage of doing away with tiny differences that are unlikely to have significantly impacted the outcomes of interest.
Median age of survival was calculated for each individual 5-year period within the data set, with changes compared between the two registries to find major points at which this value diverged. The authors then performed a multivariate analysis using the most recent interval (2009-2013) to examine whether mortality was significantly different between the two countries at a given time point when controlling for important patient baseline factors. Because not all CF is created equal, they conducted subgroup analysis based on severity of clinical phenotype (using the need for pancreatic enzyme supplementation as a proxy for worse phenotype, which apparently works) and patients specifically know to have delta F508 homozygosity.
Lastly, they performed a stratification of the US patients by insurance status, breaking them down into those with Medicaid/Medicare coverage throughout the study, those with Medicaid/Medicare for part of the study only, those with other insurance throughout the study period, and those without insurance. Interestingly, the decision to carefully track exposure to public insurance programs is underpinned by previous studies that noted worse outcomes in Medicaid patients compared to those not on Medicaid.
There clearly is a difference in median survival. After fine-tuning the registry data to minimize the risk of errors based on differential handling of patient data, the authors report a median survival in Canada of 50.9 years [95% CI, 50.5 to 52.2] compared to 40.6 years [CI, 39.1 to 41.8] in the US. Not quite the 11.7 year difference reported by the registries themselves, but still doubtless significant, even if we spot the US patients the ~2 year increase in life expectancy that the average Canadian enjoys over the US at baseline. The trend over time is interesting here:
You can see the differences start to take off at two different points, 1995 and 2005 — these points actually demonstrate significantly different rates of change on statistical analysis. It’s not clear why these years might be significant, though it’s worth noting that the US changed the way it allocates lung transplants in May 2005 (more on this later — suffice to say that it probably does not explain the observed gap). There are no clear therapeutic interventions that came out around these periods, and they are not periods in which the US was experiencing a recession. I really can’t figure it out and the authors don’t advance an explanation. Likely, these are just points when the sum of many small factors managed to tip the scales.
Interestingly, the difference here is explained preferentially by an increase in male survival. Compared to US males, survival in Canadian males increased at a significantly faster rate starting in 2005 (P < 0.001), while Canadian females did not significantly outpace American women (P = 0.19). There’s not a lot of obvious explanation for this, but it’s well-documented that women have worse outcomes than men in CF even controlling for comorbidities, perhaps because women have a tendency to be colonized with respiratory pathogens carrying a negative prognosis earlier in the disease process. Maybe the male survival simply represents their being the lower-hanging fruit, responding more rapidly to improvements in CF care.
In the multivariable model covering the period 2009-2013, Canada had a 34% lower risk of death compared to the US (HR 0.66, CI 0.54 to 0.81, P = 0.002). This association was preserved when limited to patients with a more severe clinical phenotype or those with known delta F508 homozygosity. That is a heck of an effect size. But before we start proposing a randomized controlled trial of Canadian citizenship as a possible mortality-improving intervention in CF, let’s see if we can’t pull this finding apart a bit.
First, is the finding real? The answer to that one appears to be a pretty unquestionable “yes.” The authors successfully removed systematic bias from the data analysis by effectively re-doing it all themselves. Their multivariate analysis includes the most important prognostic factors for CF, and it does not appear that the US patients were simply sicker at baseline. Even controlling for factors that could influence survival, the finding holds up.
There are more patients lost to follow up in the US database than the Canadian one, which can result in underestimation of median survival. However, this is unlikely to explain the entirety of the very wide survival gap, unless these patients uniformly lived to about 150. Given that US patients lost to follow-up were more likely to have poor prognostic factors such as female sex and pancreatic insufficiency than their Canadian counterparts, that probably did not happen.
So, why are we seeing this difference? Certainly Canadian centers adopted the high fat diet that helps to combat malnutrition in CF 5-10 years before the US, but that was back in the 1970s. That gap should be closing by now, not widening starting in 2005.
One possible is differential practices in lung transplantation. The US uses the Lung Allocation Score (LAS) to determine the order in which patients receive lung transplants, while Canada does not. This scoring system has resulted in increased transplant survival, but has resulted in fewer CF patients actually receiving a transplant (despite more being listed). Perhaps as a result, a higher proportion of US patients died without receiving a transplant, a direct expression of the different rates of transplantation (10.3% in Canada vs 6.5% in the US, P < 0.001). However, can lung transplant explain a 10 year difference in median survival? Simply, no. An intervention that only effects 10% of patients isn’t going to explain a 34% mortality advantage, and while Canadian posttransplant survival is better than in the US, median survival after transplant in Canada remains 3.3 years. So, transplant practices do not on their own seem to explain our observed differences.
That leaves one thing: insurance status. Canada has a single-payer public healthcare system, while we have our own cherished patchwork of public insurance, employer-based private plans, and a tenuous individual market via insurance exchanges (which notably did not exist during the study period). So how do patients from each segment do? Well, patients with private insurance actually did not have a significantly higher risk of death than Canadian patients (HR for Canadian patients vs US privately insured 0.85, CI 0.67 to 1.07). Canadians did do better than patients on continuous or intermittent Medicaid/Medicare, and had a 77% lower risk of death than uninsured patients in the US (HR 0.23, CI 0.14 to 0.37).
77%! It is not shocking to learn that it is important to be insured if you have a severe chronic disease, but 77% decrease in risk of death if you’re Canadian compared to an uninsured US citizen? Now this is not a prospective trial, so it’s impossible to say for sure whether the lack of insurance itself is the culprit here, or rather the low socioeconomic status that lack of insurance (or certainly Medicaid enrollment) tracks. Given that SES has been associated with poorer outcomes for CF in the past, that’s certainly possible. But the advantage of Medicaid over uninsured patients seems to make this less likely. It looks for all the world like an insurance effect.
The trial isn’t perfect. It’s out of date, with the Affordable Care Act having changed the healthcare landscape in the US (at least for the moment) since the data were collected. Further, in the interval from 2013, the last year included in the studied registries, ivacaftor/lumacaftor (Orkambi) has come to market and has the potential to revolutionize care for delta F508 homozygous CF. In clinical trials it was associated with improved PFTs and fewer pulmonary exacerbations (with an associated decrease in hospitalizations). The drug, for those who can get it (yes, we’re still talking about insurance), has the potential to be a game-changer, and so it will be important to re-examine this data at an interval after which we can see the effects of this medicine. And while investigators did a good job controlling for most known prognostic factors, retrospective cohorts can never achieve the control of confounders of a prospective RCT. But man, it’s a sobering illustration of just what our fragmented healthcare system can mean for the sickest among us.
As we as a country have discussions about what our health care system should look like, so much of it is couched in moral values about what we do or do not owe to our fellow citizens. However, one thing is clear — without universal, comprehensive access to care, patients with severe chronic disease like cystic fibrosis die sooner. Even if they have similar severity of underlying disease, even if they are prescribed similar therapeutic interventions, they die sooner. Let us not confuse the reality of what it means to have fewer people receive health care coverage, even as we discuss whether and how we ought to guarantee this.
Also: I don’t have a third Canada joke. It’s as absent as the Senators defense was in game 7 against Pittsburgh. Did I sneak it in there? Just like Chris Kunitz snuck it in on Craig Anderson. Eat it, Senators. Love, a Rangers fan.
Read the full study here.
Coming Up on IM HEAT
This Friday I will be breaking out a study of implementation of lung cancer screening in the VA system. It’s a great piece that brings the reported sensitivity, specificity, and mortality benefit of low-dose CT and translates that into what actually happens to patients when we subject them to screening. A quick taste to whet your appetite — while 56.2% of those screened had nodules requiring further investigation, only 1.5% received a diagnosis of lung cancer. More on Friday!
And hey – don’t forget to give us a follow.