• DOI:10.7326/M15-1150
• Corpus ID: 4004360

Accuracy of Peripheral Thermometers for Estimating Temperature

@article{Niven2015AccuracyOP,
  title={Accuracy of Peripheral Thermometers for Estimating Temperature},
  author={Daniel J. Niven and Jonathan E Gaudet and Kevin B. Laupland and Kelly J Mrklas and Derek J. Roberts and Henry Thomas Stelfox},
  journal={Annals of Internal Medicine},
  year={2015},
  volume={163},
  pages={768-777}
}

• D. Niven, J. Gaudet, +3 authors H. Stelfox
• Published 17 November 2015
• Medicine
• Annals of Internal Medicine

Abnormalities in body temperature (that is, fever and hypothermia) are common (14). Such abnormalities are components of diagnostic criteria for certain disorders (58), influence clinical management decisions (9, 10), and are associated with increased mortality in certain patient populations (11, 12). For most adult and pediatric populations, fever is defined as a body temperature of 38.0C or higher (1315); this definition can vary according to certain patient characteristics (for example, acute brain injury [16, 17]), institutional preference (18), and different methods of assessing body temperature (Appendix Table 1). Hypothermia is also variably defined but generally accepted to occur when body temperature is less than 36.0C (19). Appendix Table 1. Temperature Measurement Sites and Methods Accurate assessment of temperature is important. Thermometers that measure temperature from an intravascular site are considered the gold standard (20). Guidelines in adults recommend measuring temperature from one of several central options (for example, the rectum) (21), whereas guidelines in children suggest measuring from a peripheral site (for example, the axilla) (22, 23, 24). Neither recommendation is based on systematic syntheses of available evidence pertaining to thermometer accuracy (15, 25, 26). Many small studies have examined the accuracy of peripheral thermometers and report generally discordant results. Systematic reviews that examined the accuracy of peripheral thermometers in adults were either focused on selectively chosen populations or did not use established meta-analytic methods (27, 28). Meta-analyses examining infrared ear and axillary thermometry compared with a rectal reference standard in children found wide limits of agreement (LOA) (25, 26). Given these drawbacks and the potential for advancements in thermometer technology (25, 26), the accuracy of peripheral thermometers is not well-defined. We therefore did a systematic review and meta-analysis to determine whether peripheral thermometers have clinically acceptable accuracy compared with central thermometers in adults and children and whether the type of peripheral thermometer is an important determinant of accuracy. Methods The primary objective was to determine the LOA in temperature between peripheral and central thermometers (29). The secondary objective was to determine the diagnostic accuracy (sensitivity, specificity, and positive and negative likelihood ratios) of peripheral thermometers for detecting fever or hypothermia. Methods for article inclusion and data analysis were prespecified in a protocol (not registered) and consistent with major systematic review reporting guidelines (3032). Data Sources and Searches We searched MEDLINE, EMBASE, Cochrane Central Register of Controlled Trials, and CINAHL Plus for studies that compared the accuracy of peripheral thermometers with a central reference standard from inception to September 2014; we subsequently updated our search to 20 July 2015. In the MEDLINE and Cochrane Central Register of Controlled Trials searches, we used a combination of exploded Medical Subject Heading terms and text words that included synonyms for thermometer and validated filters for clinical trials (33), cohort studies (34), and diagnostic accuracy studies (Appendix Table 2) (35). The search strategy included similar combinations of terms and filters within the other databases and did not have any language restrictions. Additional citations were located using the related articles feature in PubMed and by hand-searching bibliographies of included studies, previously published reviews (15, 2528, 3640), and relevant societal guidelines (2124). Finally, 2 international clinical trial registries were searched to identify potentially relevant unpublished studies (41, 42). Appendix Table 2. MEDLINE Search Strategy* Selection Criteria Before reviewing all citations, 2 reviewers calibrated the citation screening form by reviewing 25 randomly selected citations. All titles and abstracts were then independently reviewed in duplicate. The full text of selected citations was then reviewed (in duplicate) against the inclusion and exclusion criteria. We used the following inclusion criteria: prospective study design (randomized clinical trial, controlled clinical trial, cohort study, or diagnostic accuracy study) of adults (aged 18 years) or children (aged <18 years) managed in acute care or ambulatory facilities; peripheral index thermometers included tympanic membrane, temporal artery, axillary (electronic, mercury, or chemical dot), or oral (electronic or mercury) thermometers (21, 23, 25, 26); central reference thermometers included pulmonary artery catheters or urinary bladder, esophageal, or rectal thermometers (21, 23, 25, 26); paired temperature measurements between index and reference thermometers were obtained within 5 minutes of one another; and mean difference and associated variance (SD or 95% CI) were reported for each indexreference comparison. The rectal thermometer was included as a central reference device because it is commonly viewed as invasive (22, 23, 26), frequently used in pediatric studies as the reference standard (25, 26), and listed in guidelines as being one of the most accurate thermometers (21). Exclusion criteria were as follows: casecontrol study, animal study, study population included healthy volunteers without medical concerns evaluated in nonclinical environments, or data presented in graphical format only. Noncontact infrared thermometers were excluded because they are not currently reported as a commonly used device for measuring temperature in clinical environments (18). Articles were excluded if they did not explicitly indicate that the mean difference estimates were derived using methods appropriate for repeated measures data (29). NonEnglish-language studies were translated and evaluated for inclusion. Agreement between authors at each stage of citation review was quantified using statistics. Data Extraction and Quality Assessment Data were abstracted independently and in duplicate, including the mean difference (and variance) in temperature between the index and reference thermometers and the data required to calculate the sensitivity and specificity for detecting fever and hypothermia (that is, the number of true-positive, false-positive, true-negative, and false-negative results). Data extraction that related to study quality included domains of the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) tool (43), whether the index thermometer was calibrated, and whether users received training in proper index thermometer use. Corresponding authors of selected articles were contacted for missing information as required. Data Synthesis and Analysis The primary meta-analysis determined the 95% LOA between peripheral and central thermometers; this was calculated as the pooled estimate of the mean difference in temperature measurements (peripheral minus central) 2 times the pooled estimate of the SD of the difference (29, 44). The prespecified and clinically acceptable 95% LOA were0.5C (4547). Data were pooled using a random-effects model (48), reported as the pooled mean difference (95% CI), and converted to 95% LOA. Study weights were adjusted for several indexreference comparisons. When a study used the same population to examine the accuracy of multiple index thermometers, the total study population was divided by the number of different index comparisons for pooled analyses (26); for studies that compared a given index thermometer with several reference thermometers, we restricted analyses to comparisons with the pulmonary artery catheter. The validity of pooling data across the 4 reference thermometers was examined by determining the pooled 95% LOA between each nonvascular thermometer (bladder, esophageal, or rectal) and the pulmonary artery catheter (gold standard) (21). The secondary meta-analysis determined the diagnostic accuracy of peripheral thermometers for detecting fever or hypothermia using the analytic approach outlined by the Cochrane Diagnostic Test Accuracy Working Group (32). Individual study estimates of sensitivity and specificity were determined from 22 contingency tables and used to develop the summary receiver-operating characteristic curve. Pooled estimates of sensitivity, specificity, and positive and negative likelihood ratios were then determined via bivariate random-effects models (49, 50). Interstudy heterogeneity was evaluated using the Cochran Q test and the I 2 statistic, wherein a P value less than 0.05 and an I 2 statistic greater than 25% reflected statistical evidence of interstudy heterogeneity (51). Sources of heterogeneity were examined by stratified analyses and metaregression (51). Prespecified subgroups included those pertaining to the study participants (adult vs. pediatric and medical vs. surgical), study method (high vs. low risk of bias), thermometers (index measurement site, index method of operation, index thermometer brand, user training, thermometer calibration, and type of reference standard), and temperature (fever and hypothermia). Publication bias was assessed by inspecting funnel plots and doing the Begg (primary meta-analysis) and Deek (secondary meta-analysis) funnel plot asymmetry tests (P< 0.1 was considered evidence of publication bias) (52, 53). Analyses were done in Stata, version 13.1 (Stata Corp) (54, 55). Role of the Funding Source This study received no funding. Results Study Selection and Description From 2563 citations, 75 were included in the systematic review (= 0.8 for both stages of citation review) (Appendix Figure 1). The most common reasons for exclusion after full-text review were the lack of simultaneous index and reference thermometer measurements (36%) and data pertaining to the mean difference between index and reference thermometers (34%). Sixty-nine articles were included in the primary meta