What Do Changes in Survival Rankings Tell Us About U.S. Health Care?

With funding from the Commonwealth Fund, Peter A. Muenning and Sherry A. Glied have produced a Health Affairs study of 15 year survival rates for men and women aged 45 and 65 in 13 countries for each year from 1975 to 2005.

A country’s survival rate is the percent of people of a given age who are still alive after 15 years. The authors ranked each country’s survival rate in each year. Because the U.S. relative ranking fell over the period while its per capita health spending rose, the authors conclude that the U.S. has a relatively inefficient health care system.

They blame the inefficiency on a rising number of people with “inadequate health insurance” and the reliance on “unregulated fee-for-service and specialty care” which they speculate may be “choking off public funding on more important life-saving programs.”

Little evidence is adduced to support any of these conclusions. In fact, the authors seem to jump through a myriad of speculative hoops to force the facts to conform to their preconceived views.

As is well known, age-adjusted survival rates are not by themselves a useful measure of the success or failure of a country’s health care system.

One reason that survival rates are not good measures of health system quality is that individual behavior has a large influence on health and, as a result, the health outcomes that produce survival rates. The authors do consider trans-national differences in smoking, obesity, and traffic accidents/homicide, three behavioral risk factors that receive a great deal of popular coverage. Unfortunately, their analysis of those factors is difficult to defend. At the same time, the authors ignore such factors as suicide, alcoholism, illegal drug use, poisoning, drowning, burns, and accidents other than homicide and traffic accidents.

The authors assert that “the prevalence of obesity has grown more slowly in the United States than in other nations, while smoking prevalence has declined more rapidly in the United States than in most of the comparison countries.”

Yet they provide complete obesity data only for Australia, Japan, the United Kingdom and the United States. They also note that “obesity data are based on national surveys that were not consistently collected at ten-year intervals, and data for many nations were not available or were obtained from different nationally representative surveys.”

Missing from the discussion is the possibility that obesity prevalence increased at different times in different countries.

According to Flegal et al., obesity prevalence in the United States was relatively stable from 1960-1980. It increased rapidly in the 1980s and 1990s, and may have entered a period of relative stability in the 2000s. Berghöfer et al. report that the levels of obesity seen in Europe in 2008 had already been reported in the United States 15 years earlier. If the health consequences of obesity take decades to fully manifest themselves and negatively affect survival, one would expect survival rates in the US to decline as early as 1995-2005. This is the period in which the authors’ rankings of US survival declined although US survival rates continued to improve. The health consequences of increased obesity in Europe would not yet affect European data.

Like obesity, smoking affects health some years after initiation. The authors cite Australia as the country with a level and pattern of smoking closest to that of the US and conclude that because Australia’s survival ranking did not decline, smoking is not an important contributor to the poor relative outcomes in the U.S.

In fact, US smoking prevalence likely peaked around 1974, some six years earlier than the Australian peak. US age-adjusted lung cancer incidence peaked in 1993 and remained high well into the 1995-2005 period the authors use to show U.S. decline in survival rankings. If the same time lag between smoking peak and lung cancer incidence held true in Australia, one would not expect its lung cancer mortality rate to peak before 2000. That is midway through the last ranking period used by the authors and rather late to depress what the authors call “survival gains.”

The authors declare that the share of deaths attributable to homicide and traffic accidents have been stable over time, and are “unlikely to account for the deteriorating survival probabilities of Americans.” Yet in 2004, accidents, suicides, and homicides accounted for almost 15 percent of cohort deaths in the U.S. In 1999 they accounted for slightly more than 12 percent of cohort deaths. In the early 1990s, the US death rate from accidents and violence was generally twice that of the comparison countries. Even if the US medical care had improved at the same rate as in other countries, its survival ranking would have fallen due to the fact that a higher proportion of its deaths in the age group studied would not have been affected by the hypothetical improvement in medical care.

The authors also do not discuss the possibility that different countries use different methods of classifying deaths due to traffic accidents or how those differences in classification might affect their results.

Finally, it is almost impossible to determine how the authors arrived at their survival probabilities. The paper’s survival rates are said to be calculated using “standard life table methods, in which age-specific mortality probabilities were applied to a hypothetical cohort of 100,000 people.” But age-adjusted death rates, and therefore the survival rates presented in the paper, vary with the population standard chosen, and countries use different standard populations to calculate their life tables.

In the US, the National Center for Health Statistics used the 1940 standard population from 1943 to 1998. Beginning with data year 1999, it began using the year 2000 standard population. These changes can produce changes in age-adjusted death rates that exceed the 3 percent survival variations that the authors use to rank countries. Sorlie reports that the 1995 death rate for heart diseases was 138 per 100,000 population using the 1940 standard and 296 per 100,000 using the Year 2000 standard.

Changing the population standard changes age-adjusted rates if there are substantial differences in age-specific deaths. Anderson reports that crude death rates were lower for all groups under the 2000 standard except infants and those 75 or older. Age-adjusted death rates were higher for males and lower for females.

In 2001, Ahmad et al. of the WHO examined the effect that different population standards can have on age-adjusted death rates from circulatory disease for US men. Using the Segi standard world population produces a 49.4 percent reduction in age-adjusted deaths from 1970 to 1995. Using the WHO world standard produces a 48.2 percent reduction over the same time period.

Unfortunately, the authors provide no information about the construction of the life tables used in their calculations. Readers are told that the calculations were done using Wolfram Alpha, “an online data engine.’” The authors assure the reader that it uses “rigorous, scientific tools” to calculate survival probabilities for “compiled mortality data from the WHO.”

Neither the Wolfram Alpha document nor the WHO web page cited by the authors are informative on the issue of population standards. WHO does publish a summary of the methods used in calculating its life tables for 2008, 2000, and 1990. The summary lists the data sources for the countries compared in the paper as “not available.” It is not immediately clear whether WHO changes all of its life tables when countries change their standard population, or whether its 1990 US life tables are based on the 1940 standard population while its 2005 US life tables are based on the 2000 US standard population. If the latter is the case, it obviously introduces bias into the age-adjusted mortality rates used in the paper’s calculations.

The Office for National Statistics in the United Kingdom says that the latest edition of its life tables is based on a three year standard population for 2000-2002. The 1997 life tables were based on the 1991 census. Canada’s standardized death rates for 2007, reported in February 2010, were based on a 1991 population standard.

Print