Article Text

Download PDFPDF

Original article
Modelling individual vulnerability to sexually transmitted infections to optimise intervention strategies: analysis of surveillance data from Kalamazoo County, Michigan, USA
  1. Claudio Owusu1,
  2. Kathleen M Baker2,3,
  3. Rajib Paul4,
  4. Amy B Curtis5
  1. 1 Department of Geography and Earth Sciences, University of North Carolina at Charlotte, Charlotte, North Carolina, USA
  2. 2 Department of Geography and Health Data Research, Western Michigan University, Kalamazoo, Michigan, USA
  3. 3 Health Data Research, Analysis and Mapping (HDReAM) Center, Western Michigan University, Kalamazoo, Michigan, USA
  4. 4 Department of Public Health Sciences, University of North Carolina at Charlotte, Charlotte, North Carolina, USA
  5. 5 Department of Interdisciplinary Health Sciences, Western Michigan University, Kalamazoo, Michigan, USA
  1. Correspondence to Dr Kathleen M Baker, Department of Geography and Health Data Research, Analysis and Mapping (HDReAM) Center, Western Michigan University, Kalamazoo, MI 49008, USA; kathleen.baker{at}wmich.edu

Abstract

Objective We modelled individual vulnerability to STI using personal history of infection and neighbourhood characteristics.

Methods Retrospective chlamydia and gonorrhoea data of reported confirmed cases from Kalamazoo County, Michigan for 2012 through 2014 were analysed. Unique IDs were generated from the surveillance data in collaboration with local health officials to track the individual STI histories. We then examine the concept that individuals with similar STI histories form a ‘peer’ group. These peer group include: (1) individuals with a single chlamydia; (2) individuals with single gonorrhoea; (3) individuals with repeated cases of one type of STI and (4) individuals that were diagnosed with both infections during the study period. Using Kernel density estimation, we generated densities for each peer group and assigned the intensity of the infection to the location of the individual. Finally, the individual vulnerability was characterised through ordinary least square regression (OLS) using demographics and socioeconomic variables.

Results In an OLS regression adjusted for frequency of infection, individual vulnerability to STI was only consistently significant for race and neighbourhood-level socioeconomic status (SES) in all the models under consideration. In addition, we identified six areas in three townships in Kalamazoo County that could be considered for unique interventions based on overlap patterns among peer groups.

Conclusions The results provide evidence that individual vulnerability to STI has some dependency on individual contextual (race) and exogenous factors at the neighbourhood level such as SES, regardless of that individual’s personal history of infection. We suggest place-based intervention strategies be adopted for planning STI interventions instead of current universal screening of at-risk populations.

  • peer group
  • sexually transmitted diseases
  • retrospective studies
  • gonorrhea
  • chlamydia
  • demography

Statistics from Altmetric.com

The 2014 annual surveillance report for STIs in the USA showed a significant increase in reported cases for chlamydia (2.8%) and gonorrhoea (5.1%) as well as other STIs.1 Although high reported cases of STI could certainly be due to concerted efforts being made to effectively screen for them, high numbers of individuals with repeated cases could indicate increased behavioural risk.

Many studies have identified populations who acquire an STI as being concentrated spatially in core groups.2–4 ‘Core group’ has been described as persons with repeat infections and persons with more than one sexual partner.2 5 Previous studies have found STIs, like any other communicable disease, usually spread through localised networks in neighbourhoods, because individuals interact more with their neighbours than with those further away.2 6 7 Core areas represent a continuing source of infection and are critical to STI transmission, because they account for sustained high STI rates during non-epidemic periods and act as reservoirs for infection to outside areas.8

A recent national longitudinal survey found that STI diagnosis rates were independently associated with both racial/ethnic identity and with residence in a low-income neighbourhood.9 Census tracts and other areal units have been used to define core areas and base transition rates.10 However, none of these studies investigated the different core group patterns of individuals with different STI histories.

The main objective of this analysis was to model individual vulnerability to STI using personal history of infection and neighbourhood characteristics. To achieve this objective, we examine the concept that individuals with similar STI histories form a ‘peer’ group. We modelled the density of STI cases generated by individuals in each peer group and spatially associated the appropriate peer density of cases with each individual’s address of residence. Individual characteristics and neighbourhood characteristics were examined with respect to the density of peer STI cases using ordinary least square (OLS) regression in order to better understand factors associated with individual vulnerability to STIs.

Methods

Study population

Reported, confirmed cases of chlamydia and gonorrhoea for Kalamazoo County, Michigan (figure 1) were retrieved from Michigan Diseases Surveillance System (MDSS) for 2012 through 2014. MDSS data are submitted by all healthcare providers and laboratories in Michigan for surveillance purposes. These data consistently contains the case identification number, patient name, residential address, diagnosis date, age and gender; sexual orientation and partner information were not available. Kalamazoo County, Michigan health officials identified STI management as a critical issue when chlamydia increased 10.3% and gonorrhoea 128.9% in 2014.11

Figure 1

The study area is Kalamazoo County, one of 83 counties in the state of Michigan. The county is made up of 16 townships, each 6 miles (9.5 km) on a side. Urbanised area includes portions of Kalamazoo and Portage townships.

Each STI case was assigned a unique patient identifier in the surveillance system. We developed four STI histories including: individuals with a single chlamydia case; individuals with single gonorrhoea case; individuals with multiple, separate infections of the same bacteria categorised as ‘repeaters’ and individuals, regardless of repeater status, who had a positive test for chlamydia and gonorrhoea, or ‘both’ types of infection during the study period. For individuals with multiple reports of the same infection, a case was considered new if the second positive test occurred more than 14 days after the first.2 5

Geocoding process

Residential street addresses for each individual were geocoded with ArcGIS V.10.412 commercial geographic information system (GIS) software using the 2014 street centreline obtained from the Kalamazoo County Planning Department. Batch geocoding was supplemented with extensive manual placement, resulting in an overall address match accuracy of 98.1% (n=4597) of chlamydia cases and 99.1% of gonorrhoea cases. Only cases with correctly geocoded addresses were included in further analysis.

Outcome measure

The primary outcome measure for this study was the density of cases surrounding an individual. The density of cases was a function of the count of cases within a specified search radius that were associated with the individual’s peer group. We estimated these densities after geocoding in ArcGIS V.10.4 using kernel density estimation (KDE) function.13 KDE is specified as:

Embedded Image (1)

Where n is the sample size (eg, total individuals with single chlamydia cases), ∆ is the bandwidth or the search radius, w(k) is a function evaluated at k. In the Equation (1), w(k) is evaluted at k = Embedded Image , and Embedded Image is the KDE and it is a weighted average of points near x.

KDE is a method that summarises the intensity of events across a study region13 14 within a particular bandwidth. Since STI core area can be due to local partner selection,7 a 1 km search radius was chosen as a walkability measure. KDE results, therefore, reflect the intensity of the disease at a particular location.14 15 One advantage of KDE is that it is not affected by size or shape of aggregation unit.16 We generated four different KDE results from the four STI histories being analysed in this study. The dependent variable in analysis is the intensity of cases associated with each STI peer group surrounding the locations where infection was confirmed.

Sociodemographic variables

Neighbourhoods were defined as equivalent to census tracts in this analysis because they are consistent with prior studies on neighbourhood effects in increased sexual behaviour.10 17 The American Community Survey (ACS) estimates (2010–2014) for poverty, low educational status and unemployment demographics at the census tract level were used as a measure of the neighbourhood characteristics. Poverty was used to indicate the economic hardship and has been found in related literature to be one of the most important contextual variables in health research.17 The percentage of the population with less than a high school degree was used as an indication of education with the assumption that it influences socioeconomic status (SES). In addition, lower educational attainment has also been found in related literature to be a risk factor for sexual behaviour.18 Lastly, the percentage unemployed of the population 16 years or older in the labour force was used as an indication of neighbourhood instability.

To construct an indicator for SES, we performed a principal component analysis (PCA)19 of the three census variables, including percentage of the population living below the poverty line, percentage with less than a high school education and percentage unemployed. PCA is a variable reduction procedure used to address the issue of redundancy that may cause multicollinearity between variables.19 The PCA results using extraction of components with eigenvalues >1 shows that 78.9% of the variability in the original variables was explained by the extracted component. Population density for each census tract was also used during the model development, because population density can be widely variable in urban-rural counties.

Model development

OLS multiple regression modelling was used to predict density of cases for individuals with different types of STI history. The model assumes that individuals have differences in their vulnerability (density of similar cases within 1 km) because: (1) they reside in places with varying disease prevalence and (2) neighbourhood characteristics such as SES and population density contribute to overall risk of contracting the STI. The model specification is:

Embedded Image (2)

Where ŷi is the estimated density of peer group cases around the individual, β 0represents the intercept, and β 1, β 2, β 3, β 4 are the coefficients and ԑi represents the random error term.

Because individuals had multiple diagnoses of STI, we accounted for clustering at the individual level by using the command ‘CLUSTER’ followed by the unique individual ID variable during the regression analysis in STATA V.14.2. This approach treated only cases with different personal IDs as truly independent. The advantage of using this approach is that it relaxes the assumption of independence within groups and produces robust variance estimates for the SEs.20

The density of cases was included in each regression as a continuous variable. Dummy variables were generated for categorical variables, including: gender (female, male); age (<14 years, 15–24 years, 25–34 years and >35 years) and race (black, white and other race). Age was categorised into four groups because intervention strategies are quite different for different age groups, and age was not significant when used as a continuous variable. In addition, the age groups were chosen to mirror age groups used in previous research that found that incidence and prevalence of STI varies by age groups in the USA.1 2 We categorised the PCA results for SES into two main groups of high SES (range from −1.4 to 0.5) and low SES (range from 0.6 to 3.3) after spatially exploring the results in GIS. Population density was included as a continuous variable. SES and population density were assigned to individuals based on the census tract of their residence. The baseline for comparing the categorical group coefficients in the model is male, black, 15–24 years and low SES.

Results

Between 1 January 2012 and 31 December 2014, a total of 5746 chlamydia and 1205 gonorrhoea cases were reported for Kalamazoo County. Repeat cases represented 32.3% (1856) of chlamydia and 23.3% (281) of gonorrhoea. Individuals reported up to six separate chlamydia infections and five separate gonorrhoea infections in the 3-year period.

Table 1 shows the demographic characteristics of geocoded repeaters and non-repeaters for chlamydia and gonorrhoea in the county. In addition, individuals diagnosed with both infections during the study period were included in the third portion of the table. These individuals account for 22.8% (1567) of the grand total of reporting chlamydia and gonorrhoea cases.

Table 1

Demographic characteristics of individuals with different infection history in Kalamazoo County Michigan from 2012 to 2014. In this table, individuals with both infections are represented multiple times to give an accurate assessment of the total population with a specific type of infection history

Intervention priority map

The spatial concentrations of the areas of concern for each STI history were visualised in GIS. Figure 2 shows vectorized patterns (to protect anonymity) of 20+ cases per km2 for the STI histories analysed. Areas of overlap among the distribution of peer groups indicate areas with high risky sexual behaviours. Thus, for example, a person with a single infection of STI is more likely to have multiple infections in an area with high risky sexual behaviours than if the person is in an area with no/low STI assuming persons select partners locally. These overlapping areas of high density of peer groups were considered the top priority for intervention. In total, we identified six areas in the three townships (Kalamazoo, Oshtemo and Portage) that could be considered for unique interventions based on overlap patterns among peer groups (figure 2).

Figure 2

Six areas were identified in three townships as priority areas for unique interventions based on overlapping patterns among STI peer groups.

For each cluster, the mean number of cases by type of peer group was calculated and the clusters were ranked in order of priority (figure 2). Priority was given to clusters with high density of repeat cases, high density of cases generated by individuals with both STI, high density of single gonorrhoea cases, high density of single chlamydia cases and large areas. The areas corresponding to cluster 1, 2, 3, 4 and part of 5 are in urban townships with relatively high population density (figure 2), whereas cluster 6 and an extension of cluster 5 was found in a rural township within the county (figure 2). Interestingly, all clusters vary in their contextual composition of STI history suggesting that different intervention strategies should be planned accordingly.

Statistical analysis

Results of the OLS multiple regression analysis indicated that a variety of individual-level and neighbourhood-level variables were associated with different STI histories (table 2). The model for single chlamydia case individuals shows that females with chlamydia live in areas with relatively high density of cases compared with males. In addition, whites and individuals of other race relatively live within areas with low density of single chlamydia cases compared with blacks. Individuals from 25 to 34 years and individuals 35+ years relatively live within areas with low-density cases of single chlamydia compared with individuals from 15 to 24 years.

Table 2

OLS multiple regression parameters (R2 and betas) used to model the density of individuals with varying STI histories (2012–2014) within a km2 of each individual’s residential address

The model for single gonorrhoea case individuals shows no significant differences in the density of gonorrhoea cases around the residences of female and males. In addition, non-black individuals live within areas with lower density of single gonorrhoea cases compared with blacks. No significant differences were found in the density of single gonorrhoea cases by age groups.

The model for individuals with repeat cases during the study period shows that females relatively live in areas with high density of cases compared with males. In addition, non-black individuals live within areas with lower density of individuals with repeat cases of STI compared with blacks. We also found that individuals ≤14 years significantly lived in areas with low density of individuals with repeat cases of STI compared with individuals aged 15–24 years. However, we found no significant differences in the density of repeat cases for ages 25–34 years and 35+ years compared with individuals aged 15–24 years.

For the model with individuals with diagnosed cases for both chlamydia and gonorrhoea, there was no significant differences in density for females compared with males. However, controlling for other independent variables whites and individuals of other race relatively live within areas with low density of individuals with diagnosed cases for both chlamydia and gonorrhoea compared with blacks. However, we found no significant differences in the density of single gonorrhoea cases by their age groups.

Lastly, for all the models we found that neighbourhood-level SES was significantly associated with all the STI histories. Specifically, high SES was significantly associated with a decrease in the particular STI history compared with low SES. Also, with the exception of density of single gonorrhoea cases in which population density was not significant, we recorded significance for the remaining STI models. These results therefore suggest that different factors account for vulnerability of an individual to contract STI in the community.

Discussion and conclusions

Results indicate there are significant differences in individual vulnerability to STI when considering individual factors such as race and neighbourhood-level SES. The major strength of our novel approach is the ability to model patterns in individual vulnerability to STI using personal history of infection from a retrospective data obtained in a surveillance system. To our knowledge, this is the only STI analysis that incorporates individual histories of multiple infections across disease types. Overlapping high concentration areas of confirmed STI cases were associated with different individual histories. This type of visualisation can assist community stakeholders in understanding the epidemic, educating the public and targeting interventions with limited resources. Major limitations of the study include the sensitivity of KDE to bandwidth selection during analysis; the ecological fallacy in assigning neighbourhood-level variables to individuals21 and, finally, for individuals with multiple reports of the same infection, we considered all cases with >14 days between positive tests.2 5 Future studies focused on individuals with multiple cases may choose to assign a longer waiting period between unique cases.

We found that only race and high SES were consistently significant in all models. This finding further suggests that at the individual level, blacks are more vulnerable to STI infection compared with non-blacks because they live in areas with relatively high density of STI cases. This is particularly troubling considering that only 11% of the county is black.22 In addition, we found neighbourhood high SES significantly associated with lower density of STI compared with low SES neighbourhoods. This suggest that individuals living in neighbourhoods with low SES are more vulnerable to STI compared with those living in high SES neighbourhoods. Our findings agree with results from other studies that have found STI diagnosis rates are independently associated with both racial/ethnic identity and low SES neighbourhoods.7 9

Females reported more chlamydia cases than their male counterparts; nearly three-quarters of all chlamydia repeaters were female. Females are likely to appear more frequently in the chlamydia data because of chlamydia screening recommendations—they are targeted for regular testing in the USA.23 Gonorrhoea infections were more equal in their distribution between the genders. Single gonorrhoea cases were reported slightly more often among females, and repeating cases were reported slightly more among men. These findings may reflect the fact that more males are asymptomatic for chlamydia than gonorrhoea.1 Disparities in age were fairly consistent across STI history types. At least 60 percent of single infections and over 70 percent of all repeating infections were reported in the 15–24 age groups. These results correspond to the age range found to be associated with high-risk sexual behaviours from a public health perspective.1 24 25

Based on the results, we suggest that intervention strategies such as routine STI screening for young women below the age of 24 years, and in older women with high risk of infection,23 be more focused in neighbourhoods with low SES and in areas with high proportion of minority populations. A place-based intervention is cost-effective and can also help control an outbreak of STI in communities, thereby reducing impacts at a later stage. For instance, for Kalamazoo County public health officials could setup screening programmes in the prioritised core areas of infection in addition to creating public awareness on changing risky sexual behaviours. Young adults from the ages 15–24 years had especially high numbers of STI cases in this community. We suggest a policy to allow community health officials to screen students in public schools to help reduce the burden of the infection in this population.

In conclusion, our results provide evidence that individual vulnerability to STI has some dependency on individual contextual (race) and exogenous factors at the neighbourhood level (SES) regardless of that individual’s personal history of infection. In addition, we saw that some individuals are more vulnerable to STI because they live in areas with higher numbers of STI. We therefore suggest a place-based intervention approach to reduce the burden of STI in communities. We further recommend future studies to evaluate the effectiveness of a place-based intervention as opposed to universal screening of STI.

Key messages

  • Spatial patterns in individual vulnerability to STI are associated with individual characteristics, peer group infection history and neighbourhood-level variables.

  • Race and socioeconomic status of an individual’s neighbourhood of residence are significantly associated with vulnerability to STI across all peer groups.

  • Place-based intervention strategies that are tailored to neighbourhood may be more effective than universal screening of at-risk populations.

Acknowledgments

The authors acknowledge Yasiman Back (Former Epidemiologist, Kalamazoo County Health and Community Services Department, Kalamazoo, Michigan) for her assistance with data extraction and generating unique IDs in the surveillance data.

References

Footnotes

  • Handling editor Jane S Hocking

  • Contributors CO conceived the idea that there are different patterns of core groups of STI transmission and that a multifaceted approach may be needed to target interventions to reduce the burden of STI. KMB, CO and ABC contributed in conception and research design. CO contributed in geocoding all the cases using geographic information systems (GIS). CO and KMB contributed in visualising the data in GIS and developing the interventions priority map. RP, CO and KMB contributed in developing the statistical methods used in the data analysis. ABC provided the overall supervision and coordination for the research. CO and KMB wrote the first draft of the manuscript. All authors were involved in revising the manuscript critically in order to produce an important intellectual content and the final approved version.

  • Competing interests None declared.

  • Ethics approval Western Michigan University (WMU) Human Subjects Institutional Review Board (14-09-27).

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.