This study demonstrated that different types of data entry methods may have an impact on the accuracy of patient-entered information. Within each defined category, accuracy rates associated with different data entry methods appeared to vary in relation to the complexity of requested information.
Free text entry of recalled or abstracted information proved to be a fairly accurate means of entering the names of specific diagnoses. This finding was somewhat reassuring in light of the fact that most of the PHRs in current use rely on free text entry of recalled information as a principal data entry method [25
]. It was interesting to note that subjects entering free text designations were more apt to make spelling errors in the course of entering information abstracted from clinic notes. We initially attributed these errors to illegible handwriting. Review of copies of clinic notes revealed that all but one were typewritten transcriptions of dictated entries. An alternative explanation may lie in the fact that the most of these entries included elements of medical jargon that may not be familiar to patients. This raises the question of whether diagnosis information entered as free text may need to be processed by spell-checkers that recognize acronyms and abbreviations used in clinical documentation. Subjects entering free text designations were more apt to include extraneous information that did not contribute to identification of a primary diagnosis. Most of this extraneous information focused on the assignment of etiologies or estimations of the severity of symptoms. While these modifiers did not necessarily detract from designations under consideration, their presence raised the question of whether diagnoses entered as free text may need to be parsed and sorted to isolate data elements of interest.
When entry of diagnosis information was extended to include goals of therapy, free text entry of recalled information proved to be a less accurate means of identifying principal goals of therapy. This finding was somewhat surprising in light of the fact that most of the subjects were taking prescribed thyroid hormone preparations for purposes of replacement or suppression, which are two well defined models of cause-and-effect relationships. Subjects did not fare any better in attempting to select principal goals of therapy from a categorized list of statements. The approach that focused on the selection of answers to a series of exclusionary yes/no questions proved to be the most accurate means of directing subjects to identify principal goals. This raises the question of the extent to which patients may be relied upon to directly identify their own goals of therapy. Distinction at this level may be important in situations where patients are taking agents that may be prescribed for the treatment of different conditions (e.g., diuretics, beta-blockers, systemic glucocorticoids, antiseizure medications, immunosuppressive medications). Whenever feasible, an indirect approach based on dichotomous responses to structured questions may prove to be a more reliable method of self-directed categorization.
Free text entry of recalled information was an accurate means of identifying specific names and strengths of different thyroid hormone preparations. This might have been anticipated, given the high likelihood of each subject's familiarity with this information when refilling prescriptions. For reasons that were not clear, subjects were less apt to include accurate quantitative information about units, amounts administered, and frequencies of administration in separate free text entries. This omission may have been based on the notion that this information was implicit, given the widespread use of standard dosing. It seemed less likely that this was due to lack of awareness, given that subjects following standard and nonstandard dosing regimens were able to select accurate quantitative information from pick lists. Visual identification exercises revealed that selection of tablet shapes and imprints led to more accurate identification of preparations than selection of color swatches. This discrepancy may have arisen as a result of differential browser settings, monitor settings, or variations in color perception. It should be noted that the approach based on the selection of distinctive outlines may have been successful due to the fact that all the subjects who completed this exercise were taking distinctive brand name preparations of thyroid hormone. This mode of identification may be limited in settings where the use of generic preparations that vary in shape and appearance may be more common. Direct abstraction of information from prescription labels proved to be an accurate means of entering identifying and quantitative information, irrespective of whether data elements were entered as text or selected from pick lists. Guided text entry of abstracted information might offer the advantage of greater flexibility in situations where highly variable dosing regimens may preclude generation of comprehensive pick lists (e.g., insulin regimens, immunosuppressive regi mens, adjustments of doses in chronic renal failure).
Exercises that focused on the entry of laboratory test result information suggested that the success of each approach depended in part on the source material selected for review and the degree of guidance provided in directing the abstraction of information. While subjects who engaged in free text entry of recalled information were able to identify recent tests, they were less successful in attempts to report quantitative results. Interestingly, subjects who were able to locate test results in the context of clinic notes were generally able to abstract and enter accurate qualitative and quantitative information. This exercise may have been facilitated in part by the fact that most providers documented tests of interest, results, and subsequent directives using unambiguous telegraphic styles of reporting. Approaches that rely on this mode of secondary abstraction may be confounded in situations where providers choose to document directives as annotations to laboratory test reports. Entry of a full range of qualitative, identifying, and quantitative data elements relied on directing subjects to review and abstract information from actual copies of test reports. When left to their own devices, most subjects failed to account for the source, date, units, and limits of reference ranges specified for reported results. The need for this level of detail would likely depend on the anticipated use of this information. Tracking of instances of laboratory testing might only require accurate input of source, date, and test and information. Entry of laboratory test results for purposes of disease management or self-monitoring would likely depend on accurate input of a complete range of data elements. Direct abstraction of laboratory test result information from actual copies of test reports proved to be more successful when subjects were provided with specific guidance regarding the identity, location, and format of requested data elements appearing in printed summaries. For reasons that were unclear, the only discrepancy in the accuracy of input noted was associated with the entry of unit information for requested test results. On the whole, the accuracy of guided abstraction from general format test reports appeared to match that of guided abstraction from standard commercial forms. In this case, accurate entry of information appeared to depend more on the amount of guidance provided than on the degree of constraint imposed on the range of possible entries.
The approach we adopted in designing this study had limitations. Most of the subjects we recruited were members of thyroid patient organizations and support groups. These subjects might be expected to have a certain amount of familiarity with the terminology used to describe different thyroid disorders, thyroid hormone preparations, and thyroid function tests. This may have led to overestimation of the accuracy of data entry methods. On balance, we considered this to be an acceptable risk, given some initial concerns we had about maintaining subjects' interest in participation throughout the course of the study. These concerns appeared to be borne out by the observation that a low percentage of the respondents who registered for the study actually enrolled as participants.
We chose to focus on entry of a relatively narrow range of information drawn from the domain of a particular medical subspecialty. This may have oversimplified the process of information collection by directing subjects to focus on isolated data elements. Exclusion of other diagnostic and therapeutic information may have curtailed any confusion that might have been encountered in the setting of more complex medical histories or prescription regimens.
Many of the exercises included in the study relied on the abstraction of information from documents requested directly from medical providers. While most of the subjects who were enrolled in the study were able to obtain the necessary documents with little if any difficulty, it is unclear whether this experience would be generalizable to the population at large. Given concerns about issues of liability and confidentiality, it might be reasonable to expect that patients who attempt to request documents from medical providers may encounter varying degrees of resistance. Most of the patient-oriented document organization systems in use today advocate this approach to the collection of medical information [42
Different data entry methods employed by PHRs appear to have an impact on the accuracy of patient-entered medical information. Strategic approaches adopted in planning the design of personal health records may need to take intended uses and purposes of entered information into account. Free text entry of recalled information may serve as an adequate means of entering simple designations of diagnoses, prescriptions, and laboratory tests. Accurate entry of more detailed qualitative and quantitative information may necessarily rely on approaches that prompt the guided entry of data elements abstracted from primary source documents. Further investigation should focus on evaluation of the accuracy of patient-directed entry of the full range of information that comprises a detailed medical history.