In this research, we created categorical trends in area-level deprivation from 1991 to 2006 for all of New Zealand. We then tested the relationship between these categorical trends and all-cause and CVD-related mortality rates for those areas. We found that most trends were significantly associated with the mortality outcomes, with the persistently high and persistently low trends indicating increased and decreased incidences, respectively. We also found that the only inclining trend was not significantly associated with the mortality outcomes. We found that one of the declining trends (H) was associated with significantly lower mortality than the reference moderate and stable trend, while one (F) was not. Areas in trend H tend to be rural areas, often in parts of New Zealand which have experienced increases in dairy and wine production. While the influences are likely complex, this may be an example of the middle class rising in these areas. Although the most marked decline in deprivation occurred in trend F, this was not associated with either measure of mortality. We postulate that these areas experienced financial improvement in rural and gentrification in urban settings, displacing poorer households to other areas. In addition to the decline in deprivation in trend F, these areas also exhibited lower mortality. At the beginning of the period (1991), the deprivation level in declining trend F was similar to that in trend D. At the end of the period, the deprivation level of trend F was similar to that of the reference group, making it difficult to detect a significant association. The all-cause mortality incident rate ratio (IRR) of trend D was 1.26; whereas the IRR for declining trend F was 1.09. If the lower mortality was not due to declining deprivation, we would expect similar IRRs in trends F and D.
In comparison with more ‘typical’ analyses using cross-sectional deciles of deprivation, we found that our trend measures of deprivation did not fit the model as well, as indicated by slightly higher AIC values. However, these differences between the two models were marginal, suggesting that both approaches in measuring area deprivation in relation to area-level mortality may yield similar results. Our find that using current deprivation level slightly improved model fit may relate to the primarily stable levels of deprivation over the 15-year study period. These findings could also indicate that using longitudinal measures of mortality over time may be useful in future deprivation trend analyses. Since there are a number of factors aggregated in the deprivation index, one factor could improve while another one worsens over time, yet this could result in no net change in the deprivation index value. This information about material changes could be lost when using an index in relation to health measures. In contrast to our findings, research by Riva and Curtis found that trends in area-level employment rates were (slightly) better predictors of limited long-term illness and premature mortality, measured at the individual-level, than analyses measuring area-level employment rates at one point in time . While our results did not show a major improvement in model fit of trends over deciles, we were able to observe a statistically significant, effect for the declining trend (H). Such information about health benefits of living in areas where relative deprivation has improved over time may be lost in cross-sectional analyses.
Several strengths and limitations are important to note. The primary strength of this research is its methodological contribution to the scant literature relating to longitudinal deprivation of communities and associated health impacts. In terms of potential criticism of the use of LCGM methods for inferential analysis, the defensibility of using identified trends or clusters as predictors in regression analyses has been established in other research (e.g., ). In terms of limitations specific to this study, this was an ecological study and, therefore, limitations include the inability to draw conclusions about individuals. In addition, the study does not take into account on migration of people or the length of residence in a particular place. Second, large changes in neighbourhood deprivation are rare events. Therefore, the latent class growth modelling method may not be suitable for identifying classes of rare events. To be detected, the technique may require large numbers. Stronger trends would also help elucidate relationships. Similarly, the length of the study period (15 years) may not be long enough for dramatic changes in deprivation levels to occur (e.g. processes of gentrification). The start of our study period was selected as it marks the first year that area-level deprivation (NZDep) was measured. Data for more time points, such as annually collected data, or for a longer time period would improve the current study. This approach may be most useful in urban settings to identify the processes above, or the mobility of people (e.g., poor, urban migration). Last, changes in NZDep may be due to the selection of spatial units (the modifiable area unit problem ), or changes in the actual population composition characteristics . Some caution must be used when interpreting comparisons over time. However, we have attempted to minimise these sources of error by using the census area unit level, using ranked deprivation as the measure of comparison and in examining large changes in deprivation rank only. More sophisticated work could use individual-level health outcomes and deprivation trends of both individuals and areas over the life course to aid in examining the influence of mobility on these relationships and to get closer to understanding the dynamics of disadvantage accumulation in places and in individuals. This work suggests that further work is needed to understand the potential health benefits of living in areas where relative deprivation has improved over time.