In this study we found evidence that the effect of ozone on mortality dropped off later in the ozone season (August and September). This is consistent with the hypothesis that adaptation to the acute changes associated with the mortality risk occurs, although clearly other factors may explain this pattern, which requires further study. As ozone concentrations were higher in August than in June, and only dropped off by 10% in September, this does not appear to be due to lower exposure levels. We also found evidence that the effect of ozone on mortality starts at age 50 and that the magnitude of the risk is relatively stable across most of the age range.
This pattern of change in response suggests that adaptation is not immediate, but takes several months. This is consistent with some literature. For example, one study measured lung function changes and irritant symptoms in LA residents in different seasons and, among other results, found that in fall responders had lost much of their reactivity, as if they had "adapted" to summer ambient O3 exposures. Other studies compared Los Angeles residents response to O3, to Canadians  and to new residents  and both studies showed that Los Angeles residents had a minimal response, suggesting that exposures to elevated ambient concentrations of O3 produce adaptation in residents of photochemical pollution areas.
Other studies [16, 30] showed that even if there is lung function adaptation, lung injury may persist and even cause structural damage to the respiratory tract in humans.
The biological mechanism by which ozone can affect mortality is still under examination. A review of toxicological studies found decreased heart rate, metabolism, blood pressure, and cardiac output when rats are exposed to typical concentrations of ozone .
Others studies showed that the respiratory inflammation may inhibit recovery from infection, or produce systemic responses [32–34], supporting a plausible association with cardiovascular mortality.
A study  examined whether biomarkers of inflammation are detectable in humans exposed to ozone and associated co-pollutants under natural conditions outdoors by examining 19 normal volunteer joggers with bronchoscopy with bronchoalveolar lavage (BAL). The authors found a possible ongoing inflammatory response in the lungs of recreational joggers exposed to ozone and associated co-pollutants during the summer months.
To the extent that systemic inflammation is driven by inflammation in the lung, and such lung inflammation is involved in stimulation of irritant receptors, and hence the parasympathetic responses, it is possible that the adaptation seen in previous studies of inflammatory changes could result in an attenuation of the mortality risk over the ozone season. Clearly, further work will be required to elucidate this.
Several alternative explanations need to be considered. First, a non-linear dose response relation could appear as an interaction by month if the average concentrations differ by month. Also, heat waves coincide with high ozone levels, and our control for temperature may have been inadequate to capture that. And third, there may be an interaction between ozone and temperature, which could again appear as an interaction with calendar month if the mean temperature differed by month. Regarding the first point, a recent large multi-city study by Bell and coworkers  addressed the linearity of the dose-response relation. No significant deviation was found from linearity in multiple models, and a spline model showed a lower slope only at very low ozone concentrations. The ozone concentrations were as high in August as they were in July, and almost as high in September as in May in our study. This, plus the essentially linear relation reported by Bell, makes that explanation for our findings unlikely. To address the second point we performed a sensitivity analysis where we deleted days with temperature over the 99 percentile. The results show that deleting days with high temperature did not change the results.
Finally, we do not think that in our case the different effects found by month was due to an interaction with high temperature. If there was interaction then we would expect a lower effect in May when the mean temperatures are lower, instead we found it in September, when the average temperature is higher than in May.
Then we did perform a meta-analysis dividing the cities among 6 regions and we still found that the effects in May are higher than in September in all regions (results not shown).
By season our analysis produced similar results to previous studies. For example, the meta-analyses of Levy  and Ito  reported summer effect size estimates of 0.41% and 0.39% for a 10 ppb increase in maximum hourly ozone, compared to our results, with an estimate of 0.5% for a 10 ppb increase in the 8-hour mean ozone during summer. Similarly they also didn't found an ozone effect during colder seasons.
The null effect found in winter and autumn could be explained by the lower levels of ozone in these seasons. Moreover people spend more time indoors; during colder months windows are generally closed, buildings have a lower air exchange rate and in these circumstances the levels of indoor ozone are very low.
One limitation of this study is our inability to control for ambient particles, due to the every sixth day sampling of particles. However, previous studies have reported that PM10 is not a confounder [6, 7, 37].
A recent paper  addressed the question of whether the ozone mortality relationship is confounded by secondary particles, which are produced by the same processes that produce ozone and found that the ozone effect didn't change when adjusting for PM2.5, OC or nitrate, but did decrease by 25% when adjusting for particle sulfate. Secondary particles are monitored even less than PM10, so we could not directly address the question of whether the differences we see by month could be due to different patterns of correlation by month. However, using the data from the US EPA's Speciation and Trends Network, we examined the correlations between ozone and secondary particles in all reporting cities, by month. We found the correlations between ozone and EC, OC and sulfate were lower in May compared to September. Hence, if anything, there would be more uncontrolled confounding by secondary particles in September than in May, which would be expected to inflate the September coefficient relative to May. Our study found the opposite, and hence secondary particles are unlikely to explain that finding.
This study is the first to look at fine age categories in examining the different age patterns of ozone-associated deaths. We found that the ozone-mortality association began at age fifty, and the 50–59 age group had the highest coefficient. This may be due to chance, as the effect size varied little up until age 80, after which it declined somewhat. This finding is of significance for risk assessment, since years of life lost is greater if the effect begins at age 50 than if it had no impact until after age 65.