Our findings indicate that persons who have had only one reported CD4 or VL test represent a substantial proportion of those who have a reported CD4 or VL test within 12 months of diagnosis. Estimates of the size of the population yet to be linked to care will differ under alternative definitions of linkage to care (at least one reported CD4 or VL vs
two or more). Furthermore, considering those who have had only one reported CD4 or VL test as either linked to care or alternatively, not linked to care, may mask important distinctions between persons linked to care and those not linked to care. HIV-infected persons fall along a continuum of engagement with the health care system, from persons who do not engage at all to those who engage sporadically, to those who regularly access care [33
]. If linkage to care is defined as two or more reported CD4 or VL tests, combining persons who have had only one test with those who have not had any tests potentially blurs the distinction between those who do not engage in care and those who engage sporadically. Alternatively, if linkage to care is defined as at least one reported CD4 or VL test, combining persons with only one test with those who have two or more tests potentially blurs the distinction between those who engage in care sporadically and those who regularly access care.
Additionally, we found that demographic factors independently predicted having zero reported CD4 or VL tests performed in the 12 months after diagnosis vs having only one test, but (with the exception of age at diagnosis) did not predict having only one reported test vs having two or more reported tests performed in this period. Conversely, risk group predicted having only one reported CD4 or VL vs two or more but not zero vs only one reported test(s) (with the exception of undetermined risk). These findings underscore that persons with only one reported CD4 or VL may have characteristics distinct from both those who appear to have yet to enter care and those who appear to have engaged in care. Taken together, these data suggest that a single CD4 or VL test may overestimate entry into care for certain subgroups.
Our findings support the idea that those who have received only one reported CD4 or VL test are different from those who have received more than one test. However, to better understand this difference, it is critical to know how many of the persons with a first reported CD4 or VL received their first test as part of confirmatory testing (before the first care visit). Surveillance data would miss follow-up visits during which no CD4 or VL tests are ordered. Therefore, two or more tests are needed to indicate that care visits have occurred following the confirmatory testing. Information about physicians’ practices with regard to ordering CD4 and VL tests would also be useful for evaluating linkage to care programs because it would allow assessment of the rate of uncompleted care intake visits after confirmatory testing as well as the rate at which those who have entered care drop out of care. Differentiating between care non-entry and dropping out of care is important, as different interventions are likely to be needed to overcome each of these less-than-ideal outcomes. For the reasons outlined above, all three outcomes we examined in this paper—zero reported CD4 or VL tests, only one, and two or more—should be monitored.
Although using surveillance data to track HIV care patterns appears to be feasible, there are some limitations to this approach, a primary one being that surveillance programs do not collect information on patterns of ordering CD4 and VL tests that could clarify the meaning of a single CD4 or VL test result. Another limitation is that CD4 and VL reports may be delayed, affecting the utility of these surveillance data for real-time tracking of care patterns which might help support intervention [23
]. This limitation is expected to become less important over time as jurisdictions transition to full, operational, electronic reporting of CD4 and VL results. In the meantime, retrospective analysis provides the means to track trends, albeit in a delayed timeframe.
Other factors may also affect how closely surveillance data reflect care patterns. Factors directly related to the surveillance system include incomplete reporting; receipt of medical care in a jurisdiction other than the one where HIV testing was performed; incomplete ascertainment of deaths and migration out of jurisdiction; errors such as linking reported CD4 or VL data for one person to another person’s case report; and duplication of case reports, which may happen when a person changes providers and gets re-tested for HIV [34
]. However, the surveillance system has in place procedures and policies intended to mitigate these factors, such as routine linkage of surveillance data to state and national death registries and inter-state de-duplication. Other factors not related to surveillance per se include CD4 tests ordered for a health problem unrelated to HIV, or HIV-related medical care visits that do not include ordering CD4 or VL testing. Jurisdictions must know the specific limitations of their HIV surveillance data and external factors influencing its interpretation to make optimal use of the data to monitor HIV care patterns.
Some of the limitations of surveillance data described above may be more relevant to our analyses than others. The effects of reporting delay, and of possibly incomplete reporting in jurisdictions in the early stages of implementing CD4 and VL reporting, as well as under-ascertainment of deaths and out-migration were likely minimized through retrospective analysis, which allowed more time for entry of reported CD4 and VL data and for deaths and moves out of the jurisdictions to be documented. We focused on persons diagnosed from 2006-2008, and we included CD4 and VL tests reported through July 2010. We assumed that all jurisdictions had caught up with reporting CD4 and VL of persons diagnosed in 2006-2008 by mid 2010 (reporting delays should not have affected the data presented unless the delay was longer than 18 months).
During the course of the NIC Pilot Project, incomplete and inconsistent reporting by laboratories was discovered through routine validation efforts, e.g. follow-up of inconsistent numbers of tests reported by a laboratory from month to month. Incomplete and inconsistent reporting was also discovered through the activities conducted as part of the NIC Pilot Project, e.g., when the investigation of cases or screening for eligibility indicated a person had received care, but the surveillance record included no reported CD4 or VL tests. In cases such as these, in which there was evidence that laboratory tests had indeed been ordered, the relevant laboratory was contacted and reminded of its reporting obligations. In at least one such instance, the laboratory involved was able to produce an electronic file of historical data and the missing CD4 and VL tests were added to surveillance records, and included in this analysis. Reports from these laboratories were monitored prospectively for ongoing compliance. These improvements in reporting likely benefited the Never in Care Project, having been made as they were before July 31, 2010, until which time reported CD4 or VL, regardless of when they were performed, were included in the analysis. Even so, we cannot rule out the possibility that our analysis misclassified some persons as not having a CD4 or VL test, or having only one CD4 or VL test, if CD4 or VL tests had not been reported.
It is possible that we overestimated the numbers of persons who had not received HIV medical care because reporting laws in some participating jurisdictions did not require reporting of all CD4 and all VL tests. Less frequent ordering of VL tests because of their higher cost and/or insurance coverage considerations might have contributed to overestimation of the number of persons who had not linked to HIV care. This is a concern especially for jurisdictions where CD4 counts are not reported unless they are less than 200 or less than 350 cells/mm3, because of the resulting higher probability that a care visit would be reflected by neither a reported CD4 nor a reported VL test.
We may also have overestimated the number of persons who had not linked to HIV medical care under definition 2 (i.e., when we considered those who had less than two reported CD4 or VL tests not linked to care), because persons with only one reported CD4 or VL performed on a date less than 12 months before the end of the observation period were not considered to have entered care. It is possible that the study period ended before they had time to have a second test. However, there were only 94 persons whose only reported CD4/VL test was performed less than 12 months from the end of the observation period. Therefore, even if we considered all 94 persons to have been misidentified as never in care and reclassified them, the never in care population estimate would be reduced by only 2% (from 5,257 persons to 5,163).
Finally, because we had data to estimate probabilities of not entering care only up to 42 months after HIV diagnosis, we assumed that the probability of not having entered care for persons more than 42 months from diagnosis was equal to the probability of not having entered care for persons 42 months from diagnosis. As a result, our estimates of the population having entered care were likely conservative, potentially contributing to overestimation of the numbers and percentages of persons never in care among those living with HIV (not AIDS) and diagnosed for more than 42 months in each jurisdiction.
Results from our log-binomial regression analyses indicate that jurisdiction was independently associated with not being linked to care under the two definitions, suggesting that factors other than the differences in the measured population characteristics affected the relative success in linking HIV-infected persons to care. The volume of cases to be linked to care and the resources and types of systems and services available to facilitate linkage to care varied across the jurisdictions, and these factors warrant further investigation. It is also possible that surveillance system differences may have contributed to the difference in these estimates. In addition to the differences noted above, the proportion of reported CD4 or VL results received electronically vs entered manually differed, as well as requirements for reporting of CD4 and VL tests, and length of time for which reporting of all values of CD4 or VL was mandated. Although we were unable to investigate all these factors, the sensitivity analysis results did not indicate bias in the multivariable model results from including jurisdictions with CD4 and VL reporting requirements limited to a specific numeric range. In the absence of consistent data on completeness and validity of reporting and on the availability and effectiveness of linkage to care services across the jurisdictions, we acknowledge the possibility of mixed effects of these possible influences on the findings from the multivariable models.
Despite these possible limitations, surveillance data represent the best available data for monitoring HIV care patterns on a population level. Use of surveillance data makes possible a relatively standardized assessment of linkage to care rates across an entire jurisdiction, unlike meta-analyses, for which data may not cover as wide a geographic area, for which data sources are limited to published work, and in which heterogeneity in study designs and analytic approaches can present challenges.
Our findings indicate that 85% of persons diagnosed with HIV in the five jurisdictions had a first reported CD4 or VL test performed within 12 months of diagnosis, which is substantially higher than the 69% found in a recent meta-analysis. As the meta-analysis included studies predating our observation period, a possible explanation for this difference is that linkage to care has improved over time. From our analyses, we also learned that 27% of persons with at least one reported CD4 or VL in the 12 months following diagnosis did not have a second test, suggesting that nearly one-third of those who had an initial care visit did not continue to receive regular care as recommended. Our estimate is intermediate between estimates by Ulett et al.
] and Giordano et al.
] who measured receipt of care by number of clinic visits and found that 16% and 48% of patients attending a clinic intake visit were not subsequently seen in follow-up, respectively.
In our comparison of first CD4 test results for those who had only one reported CD4 test and those who had two or more reported CD4 tests, those who had only one test were significantly more likely to have a first CD4 less than 200 cells/mm3
. This finding is consistent with other analyses of surveillance data and cohort study data indicating that many of those who do not enter or establish care soon after diagnosis delay care entry until they meet the definition of AIDS [31
Rajabiun et al.
] have reported the findings of a qualitative interview study involving an underserved population of persons living with HIV in which participants described “a tenuous connection to health care despite having seen an HIV health care provider in the recent past,” a finding that has also been reported by others [38
] and that emphasizes that efforts to improve retention in care should be implemented in parallel with efforts to engage persons living with HIV in medical care.
Our findings also emphasize that in all areas, too many HIV-infected persons are entering care too late to derive maximum benefit from antiretroviral therapy [14
]. The high proportions of persons living with HIV who had not received HIV medical care suggest that in some jurisdictions, the proportion of HIV- diagnosed persons not linked to care may exceed the proportion who have not yet been diagnosed, and that improvements in linkage to care proportionate to the expansion of HIV testing are needed. Finally, because we were focused on exploring two definitions of linkage to care, one of which did not apply to AIDS cases, we did not evaluate care utilization patterns among persons with AIDS. More work is needed to develop methods for population-level monitoring of care patterns among those with AIDS.