Optimize AIDS surveillance by sharing databases
Differing definitions, formats can cause problems
A two-year project run jointly by New York City's AIDS and tuberculosis surveillance branches found that sharing databases can optimize epidemiological information on both diseases. Success, however, depends on careful attention to format differences, confidentiality laws, and case definitions, says the director of the city's AIDS surveillance program.
"We have experienced both the pitfalls and the possibilities," says Susan Forlenza, PhD, director of the New York City Department of Health Office of AIDS Surveillance. "We have learned some lessons and developed some principles that I think others can apply."
In the past several years, the AIDS surveillance branch has used multiple disease-specific database matches to help identify new AIDS cases that are not reported to the AIDS surveillance branch and to add important information about those cases. Databases it has used include disease registries for cryptosporidiosis, tuberculosis, and cervical cancer. It also has used databases for reported deaths and CD4 count lab-based reporting.
The health department's experience with tuberculosis, however, has proven to be most challenging, Forlenza tells AIDS Alert.
"This didn't happen overnight," she says. "We had numerous conferences with the people in TB."
Differences in definitions
One complicating factor is New York state's confidentiality laws, which prohibit sharing the names of people with AIDS with any other surveil lance registry. Consequently, the sharing of information with the AIDS surveillance branch had to be a one-way street.
Secondly, different disease registries often have different standards for case definitions. Forlenza knew beforehand that the TB surveillance branch had less stringent criteria for including a definition of AIDS in its database. If the definition was too lax compared to that used for AIDS surveillance, the database might not be useful.
When 86,000 AIDS cases were compared with 36,000 TB cases, about 10,000 cases matched - those TB patients infected with HIV and who were already reported in the AIDS registry as having co-infection. However, sharing the databases enabled the AIDS surveillance branch to update TB diagnoses on about 700 cases, she notes.
At the same time, there were 524 cases in which TB patients were listed in the TB registry as having AIDS but did not show up in the AIDS registry. Such a big discrepancy forced health officials to conduct a validation study that would determine how the AIDS definitions compared between the two registries. Using a sample of 42 cases in the TB registry, health officials were able to confirm HIV diagnosis in only 79% - nearly 10% lower than the minimum the AIDS surveillance branch finds acceptable for automatically including reports in their database.
Both branches were surprised by the difference in definition criteria. "We weren't expecting such a low percentage, nor were they," Forlenza says.
One major difference was that the TB branch accepted a patient's self-reported HIV status without confirmation. The TB branch also had an HIV status category of "pending," which could have led to a patient being keyed in as HIV-positive.
As a result of the validation study, the TB branch tightened up its AIDS case definition to closely match the AIDS branch. The TB branch also had a separate, clinical database that included more follow-up information on a patient's HIV diagnosis. By incorporating that database with the first one, the case definition confirmation rate increased to 89%. A more recent validation study that used both databases and the new AIDS definition showed a confirmation rate of more than 94% - a rate good enough that the TB surveillance branch's AIDS diagnoses are now automatically entered into the AIDS registry, Forlenza says.
On the other side, the AIDS surveillance branch's definition criteria for TB are not as stringent as those used by the TB branch, Forlenza notes. For example, the AIDS database will accept acid-fast bacilli on smears and a clinical diagnosis, while the TB database requires culture confirmation. As a result, some cases that the AIDS database defines as TB have been actually turned out to be Mycobacterium avium complex. Because the TB branch uses the gold standard definition for TB, whenever a conflict in diagnosis is found, the AIDS branch automatically drops its diagnosis in deference to the TB branch's diagnosis, she adds.
Although New York law doesn't allow the AIDS branch to share its database with the TB branch, the health department provides the TB surveillance staff with epidemiological profiles of HIV-infected TB patients and cumulative data on co-infection in the city so they can better understand how the epidemics intersect.
"We have been able to show the profound intersection of the epidemics," she notes. "And in recent years we have been able to see that the actual number of cases of TB in HIV-infected people has gone down."
Cost will be too high for some databases
While the usefulness of the TB database match outweighed the complications, Forlenza recommends that health departments make a formal assessment of matches to determine whether the effort is worthwhile. For some databases, the cost will be too high, she says. As an example, Forlenza mentions an attempt to match the AIDS database with a cervical cancer database. A validation study was able to validate only 36% of cases with an AIDS diagnosis after extensive chart review and lab reports, making the database match not worthwhile, she adds.
Other state health departments have found benefits in linking computerized databases as well. The Minnesota Department of Health, for example, has linked its active AIDS surveillance database with its passive surveillance for gonorrhea, chlamydia, and syphilis to assess the incidence and risk factors for STD infection among HIV-positive people.
The results of that linkage, published in the Journal of Sexually Transmitted Diseases, have been used to help evaluate HIV/STD prevention programs, because STD incidence among HIV-positive people can be a surrogate measure of high-risk behavior in that population, the authors note.1
After matching names of 2,315 HIV-infected people in the database with the STD databases, health officials found 30 (1.3%) who were diagnosed with one or more STDs after HIV diagnosis. The age-adjusted gonorrhea incidence was 10 times higher than that of the general population, the authors note.
1. Belongia E, Danila R, Angamuthu V, et al. A population-based study of sexually transmitted disease incidence and risk factors in HIV-infected people. J Sexually Transmit Dis 1997; 22:251-256.