As our state's HIV epidemic nears the end of its third decade, the proportion of new HIV cases which are diagnosed late remains unacceptably high. Although CDC recommendations for the expansion of HIV testing have been in place for more than five years, a substantial proportion of new cases is still being detected late in the course of their HIV illness, after the point at which treatment should have been initiated. Statewide, declines in late HIV diagnosis over the past ten years appear to be minimal. Indeed, our findings support the notion that targeted HIV testing efforts, which depend heavily on patient and provider perceptions of HIV risk, cannot by themselves reduce the number of HIV-infected people who are infected but unaware of their status [3
]. Many of the characteristics we observed to be associated with late HIV diagnosis, such as being heterosexual, or residing in a rural area, are not traditionally considered strong indicators of HIV risk. Therefore, HIV testing efforts need to be broadened to include people who are at elevated risk for HIV but who are, for a variety of reasons, not getting tested.
Men in our sample were more likely to be diagnosed late than women. This finding is consistent with several published studies [24
]. However, other attempts to characterize the association between gender and late diagnosis have proven inconclusive [18
]. Using either measure of late diagnosis, the difference in the proportions of male and female cases that were diagnosed late was relatively small (<2%), suggesting the direction of the gender association could be sensitive to relatively minor differences in testing patterns. As reported elsewhere, we observed men with mode of transmission categorized as MSM or MSM/IDU to be among the least likely to be a late HIV diagnosis [28
]. These cases comprise the majority of HIV cases in Washington state. However, since these risk categories do not by definition contain female cases, they were effectively ignored by our multivariate regression model, with gender comparisons and resulting odds ratios limited mainly to cases falling into three risk categories: heterosexuals, injection drug users, and cases with no identified risk factors.
Although women are generally less likely to perceive themselves as being at risk for HIV, which would seem to favor late diagnosis, they typically demonstrate higher utilization of health services compared to men, and HIV testing is more widely accessible to women as result of both prenatal HIV screening as well as cervical cancer screening [29
]. On the other hand, the association between gender and late diagnosis could also be influenced by lower levels of HIV testing within certain male subgroups, some of whom may actually be MSM but have not been reported as such. For example, research suggests that male Latinos at risk for HIV tend to get tested for HIV less frequently, and often do not identify as being gay or bisexual despite engaging in MSM behaviors [31
We expected a positive correlation between increasing age and late diagnosis. The very act of delaying testing requires the passage of time, and as time passes, people get older. Also, as the body ages, CD4+
T-cell counts tend to naturally decrease [36
]. This results in a less-effective specific immune response, leading to greater viremia and less time to AIDS among individuals who seroconvert [37
]. Finally, some research indicates that older MSM test less often than younger MSM [39
Overall, we expected a stronger relationship between race/ethnicity and late diagnosis than we observed, given the widely documented lower testing rates among racial/ethnic minorities [16
]. While crude associations were apparent, racial/ethnic differences in late diagnosis were substantially smaller once we controlled for foreign-born status. The increased risk for late diagnosis among Native Americans is consistent with other evidence showing higher levels of poverty and limited access to health services within this population [40
]. Likewise, late diagnosis among Hispanics could be due to barriers to HIV testing, particularly stigma-induced fear of testing positive and difficulty communicating with providers among individuals who do not speak English [41
Our finding that foreign-born status confounds the associations between race/ethnicity and late HIV diagnosis is consistent with other published findings [2
]. While foreign-born cases are more likely than US-born cases to be diagnosed late, it is difficult to determine how much of this increased likelihood is due to lower HIV testing versus other reasons. As result of strict HIV surveillance reporting requirements in the US, some foreign-born cases could be diagnosed outside the US but lack required documentation to demonstrate that a previous diagnosis took place. This could result in misclassification bias, causing some foreign-born cases to appear as late diagnoses when they really are not. Needs assessments data have suggested that as many as one quarter of foreign-born HIV cases diagnosed in our state were actually diagnosed at least one year earlier than the reported date of HIV diagnosis [32
]. Yet, differences in testing behaviors could also be explained by other factors associated with recent immigration to the US, such as poor access to HIV testing services, language barriers, social isolation, financial instability, or lack of knowledge about HIV [16
4.1. Data Completeness and Limitations
Differences between cases with and without supporting data were relatively small and similar between late measures. The similarity in findings suggests that both measures would be prone to the same kinds of minimal bias.
The completeness of data supporting the time-based measure is higher than that of the lab-based measure in our study. However, this is heavily dependent on the duration of the chosen observation period. Had we evaluated late HIV diagnosis over a five-year time period, completeness would have been lower for the time-based definition.
Although we tend to assume that people with AIDS have serious health conditions that compel them to seek treatment, not all cases who develop AIDS seek care in a timely manner. Thus we may have misclassified some cases categorized as “not late” because although they have progressed to AIDS, they have not yet been linked to care or reported to our surveillance system. The time-based measure is also a subject to reporting delays associated with the diagnosis of HIV and AIDS, as well as our ability to monitor diagnoses that take place either out-of-state or outside the country.
Completeness of the data for the lab-based measure is strongly dependent on whether a given jurisdiction has implemented comprehensive lab reporting. Nevertheless, we are aware that a growing proportion of providers in our state are using out-of-state labs to process HIV-related specimens. Laws governing the reporting of laboratory results vary by state [46
]. Although Washington has laws intended to ensure the comprehensive reporting of all HIV and AIDS test results, these laws do not apply to laboratories located outside state borders. Hence, many CD4+
T-cell results likely remain unreported each year, especially if they do not correspond to an HIV or AIDS case definition, such as CD4+
T-cell counts over 200.
The logistic regression models were based on cross-sectional comparisons and do not indicate whether testing patterns have changed over time. Cases with complete data necessary to calculate either measure of lateness might not be representative of all new HIV cases, resulting in selection bias. Moderate case counts resulted in lower stratified cell sizes and wider confidence intervals that prevented our ability to detect trends in late diagnosis within stratified sub-groups.
The adjusted odds ratios generated by our logistic regression model provide an indication of which variables are associated with late HIV diagnosis in our state, as well as some idea as to relative strength of those associations. However, since late HIV diagnosis is a common event within our sample, the odds ratios likely represent a substantial overestimation of corresponding relative risks.
Among new HIV cases categorized as foreign-born in our analysis, approximately 20% are missing information about country of birth. Although some of these cases might actually have been born in the US, we do not consider this to be a large limitation. By misclassifying some native-born cases as foreign-born, we would essentially dilute the latter group, making them appear more like native-born cases than they actually are. Accurate classification of these cases would likely result in better separation between native and foreign-born cases, which would likely improve our ability to measure an association and strengthen our findings.
4.2. Strengths and Future Implications
To our knowledge, this is the first study which uses HIV surveillance data to compare two measures of late HIV diagnosis. Our results indicate that both measures point to the same risk factors for late HIV diagnosis. The relative strength and direction of these associations were also very similar.
The lab-based measure is more clinically relevant because it is based on the current recommendations for treatment initiation and can be easily modified as the standard of care evolves. In addition, it allows the inclusion of at least one additional year of data (the most recent). As our laboratory reporting system matures, and with the expected introduction of health information exchanges, the quality and completeness of laboratory data should continue to improve over time [47
]. Among the 536 cases with missing data needed to calculate the time-based measure, nearly half were identified as late using the lab-based measure. On the other hand, using both measures together could provide a broader and potentially more informative understanding of late diagnosis, offering additional information about cases that lack adequate data to support either one measure or the other.
Strengths of our study include the use of nine years of statewide surveillance data as well as an evaluation of two different ways of measuring late HIV diagnosis. Also, our use of multiple logistic regression allowed us to control for numerous potential confounders. Our results reaffirm the validity of a time-based definition of late HIV diagnosis, while at the same time demonstrating the potential value of a lab-based measure. Moreover, because it is a subject to fewer potential limitations, the lab-based measure might be a better alternative in jurisdictions with comprehensive laboratory reporting.