Incomplete Survey Data Research Articles

On calibrated inverse probability weighting and generalized boosting propensity score models for mean estimation with incomplete survey data

AbstractIncomplete data, whether realized from nonresponse in survey data or counterfactual outcomes in observational studies, may lead to biased estimation of study variables. Nonresponse and selection bias may be mitigated with techniques that weight the incomplete data to match characteristics of the partially unobserved complete data. Inverse probability weighting is a widely used method in causal inference that relies on a propensity model to construct adjusted weights; whereas calibration is a common method used by survey statisticians to use constrained optimization to construct adjusted weights. This article reviews inverse probability weighting and entropy balancing calibration by distinguishing them in the statistical sense of variable balancing, extending propensity score construction to include generalized boosting models, and demonstrating the use of inverse probability weighting and calibration separately and together through a widely cited simulation study evaluation.This article is categorized under: Statistical and Graphical Methods of Data Analysis > Sampling Data: Types and Structure > Categorical Data Statistical Learning and Exploratory Methods of the Data Sciences > Modeling Methods

144 Sports Participation and Sports-Related Neurologic Injuries in Pediatric Patients With Chiari I Malformation

INTRODUCTION: Currently, there is no consensus recommendation regarding the safety of sports participation for pediatric patients with Chiari I malformation (CM-I). METHODS: A prospectively administered survey was given to 744 pediatric neurosurgery clinic patients at the University of Michigan between December 2010 and December 2021. Data were recorded on demographic information, imaging characteristics, treatment, sports participation, and presence of sports-related neurologic injury. Patients with incomplete survey data were called. Two hundred and seven patients completed at least 1 subsequent survey and were prospectively followed for a mean of 1.56 years. A season was defined as 3 months of sports play. RESULTS: Of the 744 patients with completed surveys, 462 patients participated in sports. Patients who participated in sports were more likely to be older at presentation (p < 0.001) and have rounded cerebellar tonsil morphology (p < 0.001). A cumulative total of 6,854.67 seasons of sports were played (5,918.67 prior to CM-I decompression (CMD) and 936 after). There were 84 sports-related concussions among 55 patients; 79 prior to CMD and 5 after CMD. The concussion rate was 12.25 per 1,000 seasons of all sports, 9.16 per 1,000 seasons of limited-contact sports and 13.76 per 1,000 of contact sports. The concussion rate after CMD was 5.34 per 1,000 seasons of all sports, 9.19 per 1,000 seasons of limited-contact sports and 7.09 per 1,000 seasons of contact sports. There were no reports of long-lasting neurologic issues post-concussion or of permanent spinal cord injury. CONCLUSIONS: There were no permanent or catastrophic sports-associated neurologic injuries reported in this study. The concussion rate in CM-1 patients before and after CMD were low and therefore sports participation in this population should not be counseled against.

Reconstruction and variability of tropical pollination networks in the Brazilian Atlantic Forest

Loss of biodiversity comprehends not only the extinction of individual species, but also the loss of the ecological interactions among them. Survival of species, continuation of ecosystem functioning in nature, and ecosystem services to humans depend on the maintenance of well-functioning networks of species interactions (e.g. plant–pollinator networks and food webs). Analyses of ecological networks often rely on biased and incomplete survey data, especially in species-rich areas, such as the tropics. We used a network inference method to reconstruct pollination data compiled from a large tropical rainforest habitat extent. To gain insight into the characteristics of plant–pollinator interactions across the region, we combined the reconstructed pollination network with species distribution modelling to obtain local pollination networks throughout the area. We explored how global network properties relate to natural forest cover and land cover heterogeneity. We found that some network properties (the sum and evenness of link weights, connectance and nestedness) are positively correlated with forest cover, indicating that networks in sites with more natural habitat have greater diversity of interactions. Modularity was not related to forest cover, but seemed to reflect habitat heterogeneity, due to the broad spatial scale of the study. We believe that the methodology suggested here can facilitate the use of incomplete network data in a reliable way and allow us to better understand and protect networks of species interactions in high biodiversity regions of the world.

Evaluating Patients' Preferences Regarding Treatment Options for Waldenström's Macroglobulinemia, a Discrete-Choice-Experiment

IntroductionThe management of Waldenström's Macroglobulinemia (WM), a rare and incurable B-cell non-Hodgkin lymphoma, has evolved in recent years. Treatment options are increasing including more modern targeted therapies. Therefore there are currently several effective treatment options available for WM. There is no consensus on a preferred treatment. Widely used treatment options include rituximab combined with chemotherapy, proteasome inhibitors, and the oral BTK inhibitor ibrutinib. These treatments have varying properties in terms of efficacy, toxicity profile, duration (fixed-duration vs long-term maintenance), administration (oral vs intravenous/subcutaneous (IV/SC)), and type of agent (chemotherapy vs targeted therapy). A better understanding of patients' treatment preferences could aid physicians in developing an individualized treatment plan. Also, a better insight in patients' treatment views could help direct future clinical studies in WM. However, treatment preferences of WM patients have not been investigated. We aimed to evaluate treatment preferences of WM patients by means of a discrete choice experiment (DCE) and to assess the importance of different attributes describing the currently available treatment regimens.MethodsA mixed-method approach, consisting of a literature review, qualitative interviews and expert discussions was utilized for identification and selection of attributes/levels. A DCE questionnaire was developed in Dutch and included 5 treatment-related attributes: 5-year progression-free survival (PFS), frequency/route of administration(IV/SC or oral)/setting (clinic or home) of treatment, adverse events (nausea & vomiting and fatigue, neuropathy and atrial fibrillation), risk of future secondary malignancies (low vs high), and type of treatment agent (chemotherapy or targeted therapy). Each respondent was presented with 16 choice cards and was asked to choose between two hypothetical but realistic treatment options (see Figure 1 for an example). A pilot DCE study was carried out in 5 patients to evaluate feasibility.Data were collected via a nationwide online questionnaire via the patient organizations' website and via paper-based questionnaires sent to the participants known at the outpatient clinic. An orthogonal design was used to construct the choice tasks and a mixed logit panel data model was used to assess patients' preferences and trade-offs between attributes/levels.ResultsA total of 330 online questionnaires and 17 paper-based questionnaires were returned. After excluding incomplete survey data, 214 (65%) questionnaires were included for data analysis, respondents characteristics are presented in Table 1. The 5-year PFS followed by the risk of secondary malignancies were the most important attributes for making treatment decisions. The probability of choosing a treatment option increased with 26% and 22% if the 5-year PFS increased from 50% to 70% and if the chance of future secondary malignancies was decreased from “high risk” to “low risk”, respectively. Of the adverse events, patients disliked being at risk for neuropathy the most more than nausea, vomiting and extreme fatigue. Patients were willing to give up 7,2% treatment efficacy to avoid risk of neuropathy. The probability of choosing a treatment option increased with 8% for a fixed-duration treatment with IV/SC administration at the hospital compared to an ongoing daily oral regimen at home (Table 2). Socio-demographic characteristics such as age, gender and treatment status did not significantly influence patients' preferences with the exception of educational status. Lower 5-year PFS was more acceptable for patients with higher education (P<0.0001 ) and this subgroup of patients preferred a treatment with targeted therapy and low risk of secondary malignancy (P<0.0001).ConclusionThese are the first data on WM patients' preferences on treatment characteristics. We found that Dutch WM patients find efficacy (high 5-year PFS rate) the most important attribute, followed by a low risk of future secondary malignancies. Neuropathy was the adverse event they most wanted to avoid. They preferred a fixed-duration IV/SC treatment over an ongoing oral regimen. These data can be used in discussions with individual patients about their treatment preference, and help direct future clinical trials that optimally connect to WM patients' preferences. [Display omitted] DisclosuresMinnema: Cilag: Consultancy; Janssen: Consultancy; Celgene: Other: Travel expenses; Alnylam: Consultancy; BMS: Consultancy; Kite/Gilead: Consultancy. Kersten: Miltenyi Biotec: Consultancy, Honoraria, Other: Travel support; Roche: Consultancy, Honoraria, Other: Travel support, Research Funding; Novartis: Consultancy, Honoraria, Other: Travel support; BMS/Celgene: Consultancy, Honoraria; Takeda: Research Funding; Celgene: Research Funding; Kite, a Gilead Company: Consultancy, Honoraria, Other: Travel support, Research Funding. Vos: Celgene: Other: Travel reimbursement; Sanofi: Membership on an entity's Board of Directors or advisory committees.

Outcomes of collage art-based and narrative self-expression among home hospice caregivers

ABSTRACT Aims: This mixed methods quasi-experimental study examined the outcomes of a brief art therapy-based and a narrative interview-based intervention to address the psychosocial needs of home hospice caregivers. Methods: Participants (n = 14) chose to participate either in a narrative interview (n = 7) or arts-based interview (n = 7). Standardised self-report measures of affect, spiritual well-being, perceived stress, and self-efficacy and physiological markers of stress(salivary cortisol) and social/emotional connection (oxytocin levels) were collected before and after intervention sessions. Results: Qualitative findings indicated that caregivers (a) recognized their need for reducing stress and isolation; (b) experienced a learning curve related to illness and hospice care; (c) engaged in existential reflections on purpose and values; and were (d) living through progressive grief and loss. Quantitative findings indicated that the narrative condition group experienced improved spiritual well-being and negative affect. Both conditions were followed by improved positive affect. There were no statistically significant changes in salivary cortisol or oxytocin for either condition. Conclusion: The results suggest that caregivers might need longer term interventions for sustained changes. Implications for further research: Further research is needed to examine the role of narrative and visual self-expression among home hospice caregivers. Study limitations included a small non-randomised sample, lack of transcribed interview data, and incomplete survey and biomarker data. Plain-language summary This study examined the outcomes of an art therapy-based and a narrative intervention to address the mental, physical, and spiritual needs of family caregivers of patients at the end of life. Seven caregivers completed a therapeutic narrative interview in which caregivers shared their experiences, challenges, and sources of strength. Seven caregivers completed a collage art- based session in which they were invited to make a collage using a range of precut magazine collage images while sharing their experiences, challenges, and sources of strength as caregivers. Measures of stress (cortisol) and social/emotional connection (oxytocin) were collected before and after intervention sessions. Findings indicated that caregivers appreciated the opportunity for self-expression in both the narrative and the art-based conditions. Caregivers in the narrative condition group trended toward improvement in spiritual well-being and in reversing negative mood. Both conditions indicated a trend toward improved positive mood. Findings indicated that caregivers were consistently stressed and often isolated due to their roles. Four recurring themes were found within the qualitative data for caregivers in both conditions: (a) recognition and reduction of stress and isolation; (b) learning curve related to illness and hospice care; (c) existential reflections on death and living out values; and (d) living through progressive grief and loss. The preliminary quantitative findings suggest that narrative therapeutic approaches might be better suited for spiritual well-being. Caregivers in the narrative condition group demonstrated more improvement in spiritual well-being than those in the arts-based condition group. This finding may be the result of the format of the narrative condition in which caregivers reflected on existential concerns. In contrast, the structure of the art-making condition may have resulted in a focus on the present moment.

Friday, September 28, 2018 10:30 AM–12:00 PM abstracts: complications of cervical spine surgery: 157. Risk factors associated with failure to reach minimum clinically important difference in patient-reported outcomes after anterior cervical discectomy and fusion

BACKGROUND CONTEXT The minimum clinically importance difference (MCID) represents a threshold for improvements in patient-reported outcomes (PROs) that patients deem important following spine surgery. However, no study has comprehensively examined potential risk factors for failure to achieve MCID after anterior cervical discectomy and fusion procedures. PURPOSE To determine risk factors for failure to reach MCID for Neck Disability Index (NDI), Visual Analogue Scale (VAS) neck pain, and VAS arm pain in patients undergoing one- or two-level ACDF procedures. STUDY DESIGN/SETTING Retrospective analysis of a prospectively maintained surgical database. PATIENT SAMPLE A total of 229 patients who underwent primary, one- or two-level ACDF from 2005 to 2016. OUTCOME MEASURES NDI, CCI, ASA, VAS neck and VAS arm pain. METHODS Rates of overall MCID achievement for NDI, VAS neck pain, and VAS arm pain at final clinical follow up were calculated based on published MCID values. Bivariate testing via Poisson regression with robust error variance was used to test for association of demographic and procedural characteristics with failure to reach MCID for each PRO. The final multivariate model including all demographic and procedural categories as controls was created using backward, stepwise regression until only those variables with p RESULTS After exclusion of patients with incomplete PRO survey data, 78, 78, and 73 patients were included in the analysis for VAS neck, VAS arm, and NDI, respectively. Rates of MCID achievement for VAS neck, VAS arm, and NDI were 52.56%, 35.90%, and 73.97%, respectively. Upon bivariate analysis, patients with Charlson Comorbidity Index (CCI)≥2 were less likely to achieve MCID for NDI than patients with CCI 2 (RR 0.69, p=.017) were associated with failure to reach MCID for NDI. CONCLUSIONS The results of this study suggest that the majority of patients do not reach MCID for arm pain. Additionally, higher comorbidity burden as evidenced by higher CCI and ASA scores may be negative predictive factors for the achievement of MCID in neck disability following ACDF.

Risk Factors Associated With Failure to Reach Minimal Clinically Important Difference in Patient-reported Outcomes Following Minimally Invasive Transforaminal Lumbar Interbody Fusion for Spondylolisthesis.

Retrospective cohort. To determine risk factors associated with failure to reach the minimal clinically important difference (MCID) in patient-reported outcomes (PROs) for patients undergoing minimally invasive transforaminal lumbar interbody fusion (MIS TLIF) for spondylolisthesis. The MCID of PROs are often utilized to determine the benefit of spinal procedures. However, negative predictive factors for reaching MCID in patients surgically treated for lumbar spondylolisthesis have been difficult to elucidate. A prospectively maintained surgical database of patients who were diagnosed with lumbar spondylolisthesis and surgically treated with a single level MIS TLIF from 2010 to 2016 was reviewed. Patients with incomplete PRO survey data or <6-month follow-up were excluded from the analysis. MCID for visual analogue scale (VAS) back, VAS leg, and Oswestry Disability Index (ODI) was obtained from established values in the literature. All risk factors were then assessed for association with failure to reach MCID using bivariate and multivariate regression adjusting for preoperative characteristics. A total of 165, 76, and 73 patients treated with MIS TLIF for spondylolisthesis had complete PRO data for VAS back, VAS leg, and ODI, respectively, and were thus included in the analysis for the respective PRO. Overall, 75.76%, 71.05%, and 61.64% of patients treated with a single level MIS TLIF for spondylolisthesis reached MCID for VAS back, VAS leg, and ODI, respectively. On multivariate analysis, patients were less likely to achieve MCID for VAS back following surgical treatment if they received workers' compensation (P<0.001). No other measured factors were noted to independently correlate with MCID achievement. The results of this study suggest that a majority of patients with spondylolisthesis achieve MCID for commonly measured PROs following MIS TLIF for spondylolisthesis. However, worker's compensation insurance status may serve as a negative predictive factor for reaching MCID.

Treatment of Multivariate Outliers in Incomplete Business Survey Data

The distribution of multivariate quantitative survey data usually is not normal. Skewed and semi-continuous distributions occur often. In addition, missing values and non-response is common. All together this mix of problems makes multivariate outlier detection difficult. Examples of surveys where these problems occur are most business surveys and some household surveys like the Survey for the Statistics of Income and Living Condition (SILC) of the European Union. Several methods for multivariate outlier detection are collected in the R-package modi. This paper gives an overview of modi and its functions for outlier detection and corresponding imputation. The use of the methods is explained with a business survey dataset. The discussion covers pre- and post-processing to deal with skewness and zero-inflation, advantages and disadvantages of the methods and the choice of the parameters.

Mind the gap: What is the difference between alcohol treatment need and access for Aboriginal and Torres Strait Islander Australians?

Alcohol-related harms cause great concern to Aboriginal and Torres Strait Islander (Indigenous) communities in Australia as well as challenges to policy makers. Treatment of alcohol use disorders forms one component of an effective public health response. While alcohol dependence typically behaves as a chronic relapsing condition, treatment has been shown to be both effective and cost-effective in improving outcomes. Provision of alcohol treatment services should be based on accurate assessment of treatment need. In this paper, we examine the likely extent of the gap between voluntary alcohol treatment need and accessibility. We also suggest potential approaches to improve the ability to assess unmet need. Existing methods of assessing the treatment needs of Indigenous Australians are limited by incomplete and inaccurate survey data and an over-reliance on existing service use data. In addition to a shortage of services, cultural and logistical barriers may hamper access to alcohol treatment for Indigenous Australians. There is also a lack of services funded to a level that allows them to cope with clients with complex medical and physical comorbidity, and a lack of services for women, families and young people. A lack of voluntary treatment services also raises serious ethical concerns, given the expansion of mandatory treatment programmes and incarceration of Indigenous Australians for continued drinking. The use of modelling approaches, linkage of administrative data sets and strategies to improve data collection are discussed as possible methods to better assess treatment need. [Brett J, Lee K, Gray D, Wilson S, Freeburn B, Harrison K, Conigrave K. Mind the gap: what is the difference between alcohol treatment need and access for Aboriginal and Torres Strait Islander Australians? Drug Alcohol Rev 2016;35:456-460].

Imputing recreational angling effort from time-lapse cameras using an hierarchical Bayesian model

Digital time-lapse cameras (cameras) are increasingly used for monitoring recreational angling effort on water bodies such as lakes and rivers. Cameras are an attractive alternative to traditional methods for monitoring angling effort such as aerial counts and on-water creel surveys because of their relatively low running costs. However, cameras take photographs intermittently and it is not possible for the camera field of view to cover the entire water body in most applications. It is therefore necessary to bias correct (uprate) the camera observations of angling effort to obtain estimates of total angling effort including those out of the camera field of view. We developed a hierarchical Bayesian model to predict total angling effort from camera observations of angling effort. The model was fitted to creel effort survey data and then used to impute (‘fill-in’) total angling effort data for a larger dataset of camera observations where there were no creel survey data. The model accounted for three issues encountered when uprating camera observations of angling effort to total angling effort: (1) camera observations of zero angler effort when anglers were outside the field-of-view; (2) incomplete creel survey data; and (3) occasional data gaps caused by equipment malfunction. We applied the model to camera data from a number of small lakes in British Columbia, Canada using it to predict total angling effort that accounts for observation error. We explore the various model assumptions and discuss the limitations of the approach.

Analysis of incomplete and inconsistent clinical survey data

It is common for clinical data in survey trials to be incomplete and inconsistent for several reasons. Inconsistent data occur when more than one set of exclusive alternative questions are answered. One objective of this study was to identify and eliminate inconsistent data as an important data mining preprocessing step. We define three types of incomplete data: missing data due to skip pattern (SPMD), undetermined missing data (UMD), and genuine missing data (GMD). Identifying the type of missing data is another important objective as all missing data types cannot be treated the same. This goal cannot be achieved manually on large data of complex surveys since each subject should be processed individually. The analyses are accomplished in a mathematical framework by exploiting graph theoretic structure inherent in the questionnaire. An undirected graph is built using mutually inconsistent responses as well as its complement. The responses not in the largest maximal clique of complement graph are considered inconsistent. This guarantees removing as few responses as possible so that remaining ones are mutually consistent. Further, all potential paths in questionnaire's graph are considered, based on the responses of subjects, to identify each type of incomplete data. Experiments are conducted on MESA data. Results show 15.4 % GMD, 9.8 % SPMD, 12.9 % UMD, and 0.021 % inconsistent data. Further utility of the approach is using a) the SPMD for data stratification, and b) inconsistent data for noise estimation. Proposed method is a preprocessing prerequisite for any data mining of clinical survey data.

1-Bit matrix completion

1-Bit matrix completion

United States Private Schools Have Higher Rates of Exemptions to School Immunization Requirements than Public Schools

To compare medical, religious, and personal belief immunization exemption rates between private and public schools in US. Exemption rates were calculated using the Centers for Disease Control and Prevention School Immunization Assessment Surveys for the 2009-2010 school year excluding states with incomplete survey data. Standardized exemption rates weighted on enrollments in public and private schools were calculated. Differences in exemption rates between public and private schools were tested using Wilcoxon signed rank test. The overall state exemption rate was higher in US private than public schools, 4.25% (SD 4.27) vs 1.91% (1.67), P = .0001 and private schools had higher exemption rates for all types of exemptions; medical 0.58% (0.71) vs 0.34% (0.34) respectively (P = .0004), religious 2.09% (3.14) vs 0.83% (1.05) respectively (P = .0001), and personal belief 6.10% (4.12) vs 2.79% (1.57), respectively (P = .006). Overall exemption rates were significantly higher in states that allowed personal belief exemptions. Exemption rates were significantly higher in US private than in public schools. Children attending private schools may be at higher risk of vaccine-preventable diseases than public school children.

Imputing Missing Entries of a Data Matrix: A Review

The problem of imputing missing entries of a data matrix is easy to state: Some entries of the matrix are unknown and we want to assign appropriate values to these entries. The need for solving such problems arises in several applications, ranging from traditional ϐields to modern ones. Typical examples of traditional ϐields are statistical analysis of incomplete survey data, business reports, operations management, psychometrika, meteorology and hydrology. Modern applications arise in machine learning, data mining, DNA microarrays data, chemometrics, computer vision, recommender systems and collaborative ϐiltering. The problem is highly interesting and challenging. Many ingenious algorithms have been proposed, and there is vast literature on imputing techniques. Yet, most of the papers consider the imputing problem within the context of a speciϐic application or a speciϐic approach. The current survey provides a broad view of the problem, one that exposes the large variety of imputing methods. Old and new methods are examined with focus on the ideas behind the methods. It is illustrated how simple ideas evolve into highly sophisticated algorithms. The review enables us to identify a number of basic concepts and tools which are shared by several imputing methods. Equivalence theorems that we prove reveal surprising relations between apparently different

Predicting species distributions based on incomplete survey data: the trade‐off between precision and scale

Systematic species surveys over large areas are mostly not affordable, constraining conservation planners to make best use of incomplete data. Spatially explicit species distribution models (SDM) may be useful to detect and compensate for incomplete information. SDMs can either be based on standardized, systematic sampling in a restricted subarea, or – as a cost‐effective alternative – on data haphazardly collated by “volunteer‐based monitoring schemes” (VMS), area‐wide but inherently biased and of heterogeneous spatial precision. Using data on capercaillie Tetrao urogallus, we evaluated the capacity of SDMs generated from incomplete survey data to localise unknown areas inhabited by the species and to predict relative local observation density. Addressing the trade‐off between data precision, sample size and spatial extent of the sampling area, we compared three different sampling strategies: VMS‐data collected throughout the whole study area (7000 km2) using either 1) exact locations or 2) locations aggregated to grid cells of the size of an average individual home range, and 3) systematic transect counts conducted within a small subarea (23.8 km2). For each strategy, we compared two sample sizes and two modelling methods (ENFA and Maxent), which were evaluated using cross‐validation and independent data. Models based on VMS‐data (strategies 1 and 2) performed equally well in predicting relative observation density and in localizing “unknown” occurrences. They always outperformed strategy 3‐models, irrespective of sample size and modelling method, partly because the VMS‐data provided the more comprehensive clues for setting the discrimination‐threshold for predicting presence or absence. Accounting for potential errors due to extrapolation (e.g. projections outside the environmental domain or potentially biasing variables) reduced, but did not fully compensate for the observed discrepancies. As they cover a broader range of species‐habitat relations, the area‐wide data achieved a better model quality with less a‐priori knowledge. Furthermore, in a highly mobile species like capercaillie a sampling resolution corresponding to an individuals' home range can lead to equally good predictions as the use of exact locations. Consequently, when a trade‐off between the sampling effort and the spatial extent of the sampling area is necessary, less precise data unsystematically collected over a large representative region are preferable to systematically sampled data from a restricted region.

Mining incomplete survey data through classification

Data mining with incomplete survey data is an immature subject area. Mining a database with incomplete data, the patterns of missing data as well as the potential implication of these missing data constitute valuable knowledge. This paper presents the conceptual foundations of data mining with incomplete data through classification which is relevant to a specific decision making problem. The proposed technique generally supposes that incomplete data and complete data may come from different sub-populations. The major objective of the proposed technique is to detect the interesting patterns of data missing behavior that are relevant to a specific decision making, instead of estimation of individual missing value. Using this technique, a set of complete data is used to acquire a near-optimal classifier. This classifier provides the prediction reference information for analyzing the incomplete data. The data missing behavior concealed in the missing data is then revealed. Using a real-world survey data set, the paper demonstrates the usefulness of this technique.

Efficient estimation of population mean using incomplete survey data on study and auxiliary characteristics

This paper considers the problem of estimating the population mean using the ratio and product methods when some observations in the sample data are missing at random and the population mean of the auxiliary characteristic is not known. Besides an unbiased estimator arising from the total discard of incomplete pairs of observations, four generally biased estimators are presented. The first two estimators arise form the partial utilization of data while the remaining two are based on full utilization. A comparative study of the efficiency properties of estimators is reported and the choice of estimators is discussed.

Lessons learned from developing a tailored print intervention: a guide for practitioners and researchers new to tailoring.

Although some "how-to" guides have been written on tailored messaging, we found no reports on lessons learned from the process of developing a tailored intervention. Such lessons may be useful for practitioners and researchers who are new to tailored intervention development. The authors describe lessons gleaned from the process of developing a repeat mammography tailored print intervention. Lessons learned include the following: Selection of determinants appropriate for tailoring should be based on a theoretic framework and refined through assessment of the target population; researchers should anticipate threats to fidelity of intervention delivery because of incomplete or illogical survey data; fingerprinting enables assessment of intervention dose and how it relates to effectiveness of the tailored intervention; and a systematic process for conducting a systems test is needed to check for inconsistencies and errors before final tailored letter production. These lessons are discussed in the context of challenges and possible solutions for tailored health communication.

Classification with incomplete survey data: a Hopfield neural network approach

Survey data are often incomplete. Classification with incomplete survey data is a new subject. This study proposes a Hopfield neural network based model of classification for incomplete survey data. Using this model, an incomplete pattern is translated into fuzzy patterns. These fuzzy patterns, along with patterns without missing values, are then used as the exemplar set for teaching the Hopfield neural network. The classifier also retains information of fuzzy class membership for each exemplar pattern. When presenting a test sample, the neural network would find an exemplar that best matches the test pattern and give the classification result. Compared with other classification techniques, the proposed method can utilize more information provided by the data with missing values, and reveal the risk of the classification result on the individual observation basis.

Classification with incomplete survey data: a Hopfield neural network approach

Classification with incomplete survey data: a Hopfield neural network approach