Statistical Estimation of the Reproductive Number From Case Notification Data.

The reproductive number, or reproduction number, is a valuable metric in understanding infectious disease dynamics. There is a large body of literature related to its use and estimation. In the last 15 years, there has been tremendous progress in statistically estimating this number using case notification data. These approaches are appealing because they are relevant in an ongoing outbreak (e.g., for assessing the effectiveness of interventions) and do not require substantial modeling expertise to be implemented. In this article, we describe these methods and the extensions that have been developed. We provide insight into the distinct interpretations of the estimators proposed and provide real data examples to illustrate how they are implemented. Finally, we conclude with a discussion of available software and opportunities for future development.

Hybrid prevalence estimation: Method to improve intervention coverage estimations.

Delivering excellent health services requires accurate health information systems (HIS) data. Poor-quality data can lead to poor judgments and outcomes. Unlike probability surveys, which are representative of the population and carry accuracy estimates, HIS do not, but in many countries the HIS is the primary source of data used for administrative estimates. However, HIS are not structured to detect gaps in service coverage and leave communities exposed to unnecessary health risks. Here we propose a method to improve informatics by combining HIS and probability survey data to construct a hybrid estimator. This technique provides a more accurate estimator than either data source alone and facilitates informed decision-making. We use data from vitamin A and polio vaccination campaigns in children from Madagascar and Benin to demonstrate the effect. The hybrid estimator is a weighted average of two measurements and produces SEs and 95% confidence intervals (CIs) for the hybrid and HIS estimators. The estimates of coverage proportions using the combined data and the survey estimates differ by no more than 3%, while decreasing the SE by 1-6%; the administrative estimates from the HIS and combined data estimates are very different, with 3-25 times larger CI, questioning the value of administrative estimates. Estimators of unknown accuracy may lead to poorly formulated policies and wasted resources. The hybrid estimator technique can be applied to disease prevention services for which population coverages are measured. This methodology creates more accurate estimators, alongside measured HIS errors, to improve tracking the public's health.

Identifying treatment effects using trimmed means when data are missing not at random.

Patients often discontinue from a clinical trial because their health condition is not improving or they cannot tolerate the assigned treatment. Consequently, the observed clinical outcomes in the trial are likely better on average than if every patient had completed the trial. If these differences between trial completers and non-completers cannot be explained by the observed data, then the study outcomes are missing not at random (MNAR). One way to overcome this problem-the trimmed means approach for missing data due to study discontinuation-sets missing values as the worst observed outcome and then trims away a fraction of the distribution from each treatment arm before calculating differences in treatment efficacy (Permutt T, Li F. Trimmed means for symptom trials with dropouts. Pharm Stat. 2017;16(1):20-28). In this paper, we derive sufficient and necessary conditions for when this approach can identify the average population treatment effect. Simulation studies show the trimmed means approach's ability to effectively estimate treatment efficacy when data are MNAR and missingness due to study discontinuation is strongly associated with an unfavorable outcome, but trimmed means fail when data are missing at random. If the reasons for study discontinuation in a clinical trial are known, analysts can improve estimates with a combination of multiple imputation and the trimmed means approach when the assumptions of each hold. We compare the methodology to existing approaches using data from a clinical trial for chronic pain. An R package trim implements the method. When the assumptions are justifiable, using trimmed means can help identify treatment effects notwithstanding MNAR data.

Role of survey response rates on valid inference: an application to HIV prevalence estimates.

BACKGROUND: Nationally-representative surveys suggest that females have a higher prevalence of HIV than males in most African countries. Unfortunately, these results are made on the basis of surveys with non-ignorable missing data. This study evaluates the impact that differential survey nonresponse rates between males and females can have on the point estimate of the HIV prevalence ratio of these two classifiers. METHODS: We study 29 Demographic and Health Surveys (DHS) from 2001 to 2010. Instead of employing often used multiple imputation models with a Missing at Random assumption that may not hold in this setting, we assess the effect of ignoring the information contained in the missing HIV information for males and females through three proposed statistical measures. These measures can be used in settings where the interest is comparing the prevalence of a disease between two groups. The proposed measures do not utilize parametric models and can be implemented by researchers of any level. They are: (1) an upper bound on the potential bias of the usual practise of using reported HIV prevalence estimates that ignore subjects who have missing HIV outcomes. (2) Plausible range intervals to account for nonresponses, without any additional parametric modeling assumptions. (3) Prevalence ratio inflation factors to correct the point estimate of the HIV prevalence ratio, if estimates of nonresponders' HIV prevalences were known. RESULTS: In 86% of countries, males have higher upper bounds of HIV prevalence than females, this is consonant with males possibly having higher infection rates than females. Additionally, 74% of surveys have a plausible range that crosses 1.0, suggesting a plausible equivalence between male and female HIV prevalences. CONCLUSIONS: It is quite reasonable to conclude that there is so much DHS nonresponse in evaluating the HIV status question, that existing data is plausibly generated by the situation where the virus is equally distributed between the sexes.

Utilizing the Jaccard index to reveal population stratification in sequencing data: a simulation study and an application to the 1000 Genomes Project.

MOTIVATION: Population stratification is one of the major sources of confounding in genetic association studies, potentially causing false-positive and false-negative results. Here, we present a novel approach for the identification of population substructure in high-density genotyping data/next generation sequencing data. The approach exploits the co-appearances of rare genetic variants in individuals. The method can be applied to all available genetic loci and is computationally fast. Using sequencing data from the 1000 Genomes Project, the features of the approach are illustrated and compared to existing methodology (i.e. EIGENSTRAT). We examine the effects of different cutoffs for the minor allele frequency on the performance of the approach. We find that our approach works particularly well for genetic loci with very small minor allele frequencies. The results suggest that the inclusion of rare-variant data/sequencing data in our approach provides a much higher resolution picture of population substructure than it can be obtained with existing methodology. Furthermore, in simulation studies, we find scenarios where our method was able to control the type 1 error more precisely and showed higher power. CONTACT: dmitry.prokopenko@uni-bonn.de SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Multidrug-resistant tuberculosis treatment failure detection depends on monitoring interval and microbiological method.

Debate persists about monitoring method (culture or smear) and interval (monthly or less frequently) during treatment for multidrug-resistant tuberculosis (MDR-TB). We analysed existing data and estimated the effect of monitoring strategies on timing of failure detection.We identified studies reporting microbiological response to MDR-TB treatment and solicited individual patient data from authors. Frailty survival models were used to estimate pooled relative risk of failure detection in the last 12 months of treatment; hazard of failure using monthly culture was the reference.Data were obtained for 5410 patients across 12 observational studies. During the last 12 months of treatment, failure detection occurred in a median of 3 months by monthly culture; failure detection was delayed by 2, 7, and 9 months relying on bimonthly culture, monthly smear and bimonthly smear, respectively. Risk (95% CI) of failure detection delay resulting from monthly smear relative to culture is 0.38 (0.34-0.42) for all patients and 0.33 (0.25-0.42) for HIV-co-infected patients.Failure detection is delayed by reducing the sensitivity and frequency of the monitoring method. Monthly monitoring of sputum cultures from patients receiving MDR-TB treatment is recommended. Expanded laboratory capacity is needed for high-quality culture, and for smear microscopy and rapid molecular tests.

Evaluation of high-dose rifampin in patients with new, smear-positive tuberculosis (HIRIF): study protocol for a randomized controlled trial.

BACKGROUND: Evidence has existed for decades that higher doses of rifampin may be more effective, but potentially more toxic, than standard doses used in tuberculosis treatment. Whether increased doses of rifampin could safely shorten treatment remains an open question. METHODS/DESIGN: The HIRIF study is a phase II randomized trial comparing rifampin doses of 20 and 15 mg/kg/day to the standard 10 mg/kg/day for the first 2 months of tuberculosis treatment. All participants receive standard doses of companion drugs and a standard continuation-phase treatment (4 months, 2 drugs). They are followed for 6 months post treatment. Study participants are adults with newly diagnosed, previously untreated, smear positive (≥2+) pulmonary tuberculosis. The primary outcome is rifampin area under the plasma concentration-time curve (AUC0-24) after at least 14 days of study treatment/minimum inhibitory concentration. 180 randomized participants affords 90 % statistical power to detect a difference of at least 14 mcg/mL*hr between the 20 mg/kg group and the 10 mg/kg group, assuming a loss to follow-up of up to 17 %. DISCUSSION: Extant evidence suggests the potential for increased doses of rifampin to shorten tuberculosis treatment duration. Early studies that explored this potential using intermittent, higher dosing were derailed by toxicity. Given the continued large, global burden of tuberculosis with nearly 10 million new cases annually, shortened regimens with existing drugs would offer an important advantage to patients and health systems. TRIAL REGISTRATION: This trial was registered with clinicaltrials.gov (registration number: NCT01408914 ) on 2 August 2011.

Institutional Tuberculosis Transmission. Controlled Trial of Upper Room Ultraviolet Air Disinfection: A Basis for New Dosing Guidelines.

RATIONALE: Transmission is driving the global tuberculosis epidemic, especially in congregate settings. Worldwide, natural ventilation is the most common means of air disinfection, but it is inherently unreliable and of limited use in cold climates. Upper room germicidal ultraviolet (UV) air disinfection with air mixing has been shown to be highly effective, but improved evidence-based dosing guidelines are needed. OBJECTIVES: To test the efficacy of upper room germicidal air disinfection with air mixing to reduce tuberculosis transmission under real hospital conditions, and to define the application parameters responsible as a basis for proposed new dosing guidelines. METHODS: Over an exposure period of 7 months, 90 guinea pigs breathed only untreated exhaust ward air, and another 90 guinea pigs breathed only air from the same six-bed tuberculosis ward on alternate days when upper room germicidal air disinfection was turned on throughout the ward. MEASUREMENTS AND MAIN RESULTS: The tuberculin skin test conversion rates (>6 mm) of the two chambers were compared. The hazard ratio for guinea pigs in the control chamber converting their skin test to positive was 4.9 (95% confidence interval, 2.8-8.6), with an efficacy of approximately 80%. CONCLUSIONS: Upper room germicidal UV air disinfection with air mixing was highly effective in reducing tuberculosis transmission under hospital conditions. These data support using either a total fixture output (rather than electrical or UV lamp wattage) of 15-20 mW/m(3) total room volume, or an average whole-room UV irradiance (fluence rate) of 5-7 µW/cm(2), calculated by a lighting computer-assisted design program modified for UV use.

Incidence of multidrug-resistant tuberculosis disease in children: systematic review and global estimates.

BACKGROUND: Multidrug-resistant tuberculosis threatens to reverse recent reductions in global tuberculosis incidence. Although children younger than 15 years constitute more than 25% of the worldwide population, the global incidence of multidrug-resistant tuberculosis disease in children has never been quantified. We aimed to estimate the regional and global annual incidence of multidrug-resistant tuberculosis in children. METHODS: We developed two models: one to estimate the setting-specific risk of multidrug-resistant tuberculosis among child cases of tuberculosis, and a second to estimate the setting-specific incidence of tuberculosis disease in children. The model for risk of multidrug-resistant tuberculosis among children with tuberculosis needed a systematic literature review. We multiplied the setting-specific estimates of multidrug-resistant tuberculosis risk and tuberculosis incidence to estimate regional and global incidence of multidrug-resistant tuberculosis disease in children in 2010. FINDINGS: We identified 3403 papers, of which 97 studies met inclusion criteria for the systematic review of risk of multidrug-resistant tuberculosis. 31 studies reported the risk of multidrug-resistant tuberculosis in both children and treatment-naive adults with tuberculosis and were used for evaluation of the linear association between multidrug-resistant disease risk in these two patient groups. We identified that the setting-specific risk of multidrug-resistant tuberculosis was nearly identical in children and treatment-naive adults with tuberculosis, consistent with the assertion that multidrug-resistant disease in both groups reflects the local risk of transmitted multidrug-resistant tuberculosis. After application of these calculated risks, we estimated that around 999,792 (95% CI 937,877-1,055,414) children developed tuberculosis disease in 2010, of whom 31,948 (25,594-38,663) had multidrug-resistant disease. INTERPRETATION: Our estimates underscore that many cases of tuberculosis and multidrug-resistant tuberculosis disease are not being detected in children. Future estimates can be refined as more and better tuberculosis data and new diagnostic instruments become available. FUNDING: US National Institutes of Health, the Helmut Wolfgang Schumann Fellowship in Preventive Medicine at Harvard Medical School, the Norman E Zinberg Fellowship at Harvard Medical School, and the Doris and Howard Hiatt Residency in Global Health Equity and Internal Medicine at the Brigham and Women's Hospital.

Comparing two survey methods of measuring health-related indicators: Lot Quality Assurance Sampling and Demographic Health Surveys.

OBJECTIVES: Two common methods used to measure indicators for health programme monitoring and evaluation are the demographic and health surveys (DHS) and lot quality assurance sampling (LQAS); each one has different strengths. We report on both methods when utilised in comparable situations. METHODS: We compared 24 indicators in south-west Uganda, where data for prevalence estimations were collected independently for the two methods in 2011 (LQAS: n = 8876; DHS: n = 1200). Data were stratified (e.g. gender and age) resulting in 37 comparisons. We used a two-sample two-sided Z-test of proportions to compare both methods. RESULTS: The average difference between LQAS and DHS for 37 estimates was 0.062 (SD = 0.093; median = 0.039). The average difference among the 21 failures to reject equality of proportions was 0.010 (SD = 0.041; median = 0.009); among the 16 rejections, it was 0.130 (SD = 0.010, median = 0.118). Seven of the 16 rejections exhibited absolute differences of <0.10, which are clinically (or managerially) not significant; 5 had differences >0.10 and <0.20 (mean = 0.137, SD = 0.031) and four differences were >0.20 (mean = 0.261, SD = 0.083). CONCLUSION: There is 75.7% agreement across the two surveys. Both methods yield regional results, but only LQAS provides information at less granular levels (e.g. the district level) where managerial action is taken. The cost advantage and localisation make LQAS feasible to conduct more frequently, and provides the possibility for real-time health outcomes monitoring.

Estimation after classification using lot quality assurance sampling: corrections for curtailed sampling with application to evaluating polio vaccination campaigns.

OBJECTIVES: To assess the bias incurred when curtailment of Lot Quality Assurance Sampling (LQAS) is ignored, to present unbiased estimators, to consider the impact of cluster sampling by simulation and to apply our method to published polio immunization data from Nigeria. METHODS: We present estimators of coverage when using two kinds of curtailed LQAS strategies: semicurtailed and curtailed. We study the proposed estimators with independent and clustered data using three field-tested LQAS designs for assessing polio vaccination coverage, with samples of size 60 and decision rules of 9, 21 and 33, and compare them to biased maximum likelihood estimators. Lastly, we present estimates of polio vaccination coverage from previously published data in 20 local government authorities (LGAs) from five Nigerian states. RESULTS: Simulations illustrate substantial bias if one ignores the curtailed sampling design. Proposed estimators show no bias. Clustering does not affect the bias of these estimators. Across simulations, standard errors show signs of inflation as clustering increases. Neither sampling strategy nor LQAS design influences estimates of polio vaccination coverage in 20 Nigerian LGAs. When coverage is low, semicurtailed LQAS strategies considerably reduces the sample size required to make a decision. Curtailed LQAS designs further reduce the sample size when coverage is high. CONCLUSIONS: Results presented dispel the misconception that curtailed LQAS data are unsuitable for estimation. These findings augment the utility of LQAS as a tool for monitoring vaccination efforts by demonstrating that unbiased estimation using curtailed designs is not only possible but these designs also reduce the sample size.

Novel developmental analyses identify longitudinal patterns of early gut microbiota that affect infant growth.

It is acknowledged that some obesity trajectories are set early in life, and that rapid weight gain in infancy is a risk factor for later development of obesity. Identifying modifiable factors associated with early rapid weight gain is a prerequisite for curtailing the growing worldwide obesity epidemic. Recently, much attention has been given to findings indicating that gut microbiota may play a role in obesity development. We aim at identifying how the development of early gut microbiota is associated with expected infant growth. We developed a novel procedure that allows for the identification of longitudinal gut microbiota patterns (corresponding to the gut ecosystem developing), which are associated with an outcome of interest, while appropriately controlling for the false discovery rate. Our method identified developmental pathways of Staphylococcus species and Escherichia coli that were associated with expected growth, and traditional methods indicated that the detection of Bacteroides species at day 30 was associated with growth. Our method should have wide future applicability for studying gut microbiota, and is particularly important for translational considerations, as it is critical to understand the timing of microbiome transitions prior to attempting to manipulate gut microbiota in early life.

Extending cluster lot quality assurance sampling designs for surveillance programs.

Lot quality assurance sampling (LQAS) has a long history of applications in industrial quality control. LQAS is frequently used for rapid surveillance in global health settings, with areas classified as poor or acceptable performance on the basis of the binary classification of an indicator. Historically, LQAS surveys have relied on simple random samples from the population; however, implementing two-stage cluster designs for surveillance sampling is often more cost-effective than simple random sampling. By applying survey sampling results to the binary classification procedure, we develop a simple and flexible nonparametric procedure to incorporate clustering effects into the LQAS sample design to appropriately inflate the sample size, accommodating finite numbers of clusters in the population when relevant. We use this framework to then discuss principled selection of survey design parameters in longitudinal surveillance programs. We apply this framework to design surveys to detect rises in malnutrition prevalence in nutrition surveillance programs in Kenya and South Sudan, accounting for clustering within villages. By combining historical information with data from previous surveys, we design surveys to detect spikes in the childhood malnutrition rate.

Determining the dynamics of influenza transmission by age.

BACKGROUND: It is widely accepted that influenza transmission dynamics vary by age; however methods to quantify the reproductive number by age group are limited. We introduce a simple method to estimate the reproductive number by modifying the method originally proposed by Wallinga and Teunis and using existing information on contact patterns between age groups. We additionally perform a sensitivity analysis to determine the potential impact of differential healthcare seeking patterns by age. We illustrate this method using data from the 2009 H1N1 Influenza pandemic in Gauteng Province, South Africa. RESULTS: Our results are consistent with others in showing decreased transmission with age. We show that results can change markedly when we make the account for differential healthcare seeking behaviors by age. CONCLUSIONS: We show substantial heterogeneity in transmission by age group during the Influenza A H1N1 pandemic in South Africa. This information can greatly assist in targeting interventions and implementing social distancing measures.

The effect of spatial aggregation on performance when mapping a risk of disease.

BACKGROUND: Spatial data on cases are available either in point form (e.g. longitude/latitude), or aggregated by an administrative region (e.g. zip code or census tract). Statistical methods for spatial data may accommodate either form of data, however the spatial aggregation can affect their performance. Previous work has studied the effect of spatial aggregation on cluster detection methods. Here we consider geographic health data at different levels of spatial resolution, to study the effect of spatial aggregation on disease mapping performance in locating subregions of increased disease risk. METHODS: We implemented a non-parametric disease distance-based mapping (DBM) method to produce a smooth map from spatially aggregated childhood leukaemia data. We then simulated spatial data under controlled conditions to study the effect of spatial aggregation on its performance. We used an evaluation method based on ROC curves to compare performance of DBM across different geographic scales. RESULTS: Application of DBM to the leukaemia data illustrates the method as a useful visualization tool. Spatial aggregation produced expected degradation of disease mapping performance. Characteristics of this degradation, however, varied depending on the interaction between the geographic extent of the higher risk area and the level of aggregation. For example, higher risk areas dispersed across several units did not suffer as greatly from aggregation. The choice of centroids also had an impact on the resulting mapping. CONCLUSIONS: DBM can be implemented for continuous and discrete spatial data, but the resulting mapping can lose accuracy in the second setting. Investigation of the simulations suggests a complex relationship between performance loss, geographic extent of spatial disturbances and centroid locations. Aggregation of spatial data destroys information and thus impedes efforts to monitor these data for spatial disturbances. The effect of spatial aggregation on cluster detection, disease mapping, and other useful methods in spatial epidemiology is complex and deserves further study.

'Location, Location, Location': a spatial approach for rare variant analysis and an application to a study on non-syndromic cleft lip with or without cleft palate.

MOTIVATION: For the analysis of rare variants in sequence data, numerous approaches have been suggested. Fixed and flexible threshold approaches collapse the rare variant information of a genomic region into a test statistic with reduced dimensionality. Alternatively, the rare variant information can be combined in statistical frameworks that are based on suitable regression models, machine learning, etc. Although the existing approaches provide powerful tests that can incorporate information on allele frequencies and prior biological knowledge, differences in the spatial clustering of rare variants between cases and controls cannot be incorporated. Based on the assumption that deleterious variants and protective variants cluster or occur in different parts of the genomic region of interest, we propose a testing strategy for rare variants that builds on spatial cluster methodology and that guides the identification of the biological relevant segments of the region. Our approach does not require any assumption about the directions of the genetic effects. RESULTS: In simulation studies, we assess the power of the clustering approach and compare it with existing methodology. Our simulation results suggest that the clustering approach for rare variants is well powered, even in situations that are ideal for standard methods. The efficiency of our spatial clustering approach is not affected by the presence of rare variants that have opposite effect size directions. An application to a sequencing study for non-syndromic cleft lip with or without cleft palate (NSCL/P) demonstrates its practical relevance. The proposed testing strategy is applied to a genomic region on chromosome 15q13.3 that was implicated in NSCL/P etiology in a previous genome-wide association study, and its results are compared with standard approaches. AVAILABILITY: Source code and documentation for the implementation in R will be provided online. Currently, the R-implementation only supports genotype data. We currently are working on an extension for VCF files. CONTACT: heide.fier@googlemail.com.

Estimating HIV prevalence from surveys with low individual consent rates: annealing individual and pooled samples.

: Many HIV prevalence surveys are plagued by the problem that a sizeable number of surveyed individuals do not consent to contribute blood samples for testing. One can ignore this problem, as is often done, but the resultant bias can be of sufficient magnitude to invalidate the results of the survey, especially if the number of non-responders is high and the reason for refusing to participate is related to the individual's HIV status. One reason for refusing to participate may be for reasons of privacy. For those individuals, we suggest offering the option of being tested in a pool. This form of testing is less certain than individual testing, but, if it convinces more people to submit to testing, it should reduce the potential for bias and give a cleaner answer to the question of prevalence. This paper explores the logistics of implementing a combined individual and pooled testing approach and evaluates the analytical advantages to such a combined testing strategy. We quantify improvements in a prevalence estimator based on this combined testing strategy, relative to an individual testing only approach and a pooled testing only approach. Minimizing non-response is key for reducing bias, and, if pooled testing assuages privacy concerns, offering a pooled testing strategy has the potential to substantially improve HIV prevalence estimates.

The effect of clustering on lot quality assurance sampling: a probabilistic model to calculate sample sizes for quality assessments.

BACKGROUND: Traditional Lot Quality Assurance Sampling (LQAS) designs assume observations are collected using simple random sampling. Alternatively, randomly sampling clusters of observations and then individuals within clusters reduces costs but decreases the precision of the classifications. In this paper, we develop a general framework for designing the cluster(C)-LQAS system and illustrate the method with the design of data quality assessments for the community health worker program in Rwanda. RESULTS: To determine sample size and decision rules for C-LQAS, we use the beta-binomial distribution to account for inflated risk of errors introduced by sampling clusters at the first stage. We present general theory and code for sample size calculations.The C-LQAS sample sizes provided in this paper constrain misclassification risks below user-specified limits. Multiple C-LQAS systems meet the specified risk requirements, but numerous considerations, including per-cluster versus per-individual sampling costs, help identify optimal systems for distinct applications. CONCLUSIONS: We show the utility of C-LQAS for data quality assessments, but the method generalizes to numerous applications. This paper provides the necessary technical detail and supplemental code to support the design of C-LQAS for specific programs.

Estimating the reproductive number in the presence of spatial heterogeneity of transmission patterns.

BACKGROUND: Estimates of parameters for disease transmission in large-scale infectious disease outbreaks are often obtained to represent large groups of people, providing an average over a potentially very diverse area. For control measures to be more effective, a measure of the heterogeneity of the parameters is desirable. METHODS: We propose a novel extension of a network-based approach to estimating the reproductive number. With this we can incorporate spatial and/or demographic information through a similarity matrix. We apply this to the 2009 Influenza pandemic in South Africa to understand the spatial variability across provinces. We explore the use of five similarity matrices to illustrate their impact on the subsequent epidemic parameter estimates. RESULTS: When treating South Africa as a single entity with homogeneous transmission characteristics across the country, the basic reproductive number, R0, (and imputation range) is 1.33 (1.31, 1.36). When fitting a new model for each province with no inter-province connections this estimate varies little (1.23-1.37). Using the proposed method with any of the four similarity measures yields an overall R0 that varies little across the four new models (1.33 to 1.34). However, when allowed to vary across provinces, the estimated R0 is greater than one consistently in only two of the nine provinces, the most densely populated provinces of Gauteng and Western Cape. CONCLUSIONS: Our results suggest that the spatial heterogeneity of influenza transmission was compelling in South Africa during the 2009 pandemic. This variability makes a qualitative difference in our understanding of the epidemic. While the cause of this fluctuation might be partially due to reporting differences, there is substantial evidence to warrant further investigation.