Trends and Between-Physician Variation in Laboratory Testing: A Retrospective Longitudinal Study in General Practice

Abstract

Laboratory tests are frequently ordered by general practitioners (GPs), but little is known about time trends and between-GP variation of their use. In this retrospective longitudinal study, we analyzed over six million consultations by Swiss GPs during the decade 2009-2018. For 15 commonly used test types, we defined specific laboratory testing rates (sLTR) as the percentage of consultations involving corresponding laboratory testing requests. Patient age- and sex-adjusted time trends of sLTR were modeled with mixed-effect logistic regression accounting for clustering of patients within GPs. We quantified between-GP variation by means of intraclass correlation coefficients (ICC). Nine out of the 15 laboratory test types considered showed significant temporal increases, most eminently vitamin D (ten-year odds ratio (OR) 1.88, 95% confidence interval (CI) 1.71-2.06) and glycated hemoglobin (ten-year OR 1.87, 95% CI 1.82-1.92). Test types both subject to substantial increase and high between-GP variation of sLTR were vitamin D (ICC 0.075), glycated hemoglobin (ICC 0.101), C-reactive protein (ICC 0.202), and vitamin B12 (ICC 0.166). Increasing testing frequencies and large between-GP variation of specific test type use pointed at inconsistencies of medical practice and potential overuse.

Keywords: general practice; intraclass correlation coefficient; laboratory testing; mixed-effect model; trend.

Figure 1

Crude among-general practitioner (GP) distributions of 2009–2018 average specific laboratory testing rates. Test types were included according to the criteria described in the main text. Type-specific laboratory testing rates were calculated for each GP as the percentage of consultations during the GP’s observation period involving a request of the respective test type. Abbreviations: CBC, complete blood count; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; HbA1c, glycated hemoglobin; PT/INR, prothrombin time/international normalized ratio; TSH, thyroid-stimulating hormone.

Figure 2

Crude among-general practitioner (GP) distributions of annual average specific laboratory testing rates for the years 2009–2018. Test types were included according to the criteria described in the main text. Outliers are omitted for better readability. Abbreviations: CBC, complete blood count; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; HbA1c, glycated hemoglobin; PT/INR, prothrombin time/international normalized ratio; TSH, thyroid-stimulating hormone.

Figure 3

Results of mixed-effect regression analysis for specific laboratory testing rates. (a) Ten-year time trends. (b) Effect sizes of patient age. (c) Effect sizes of patient sex. (d) Between-general practitioner variance in terms of the null-model intraclass correlation coefficient. Abbreviations: CBC, complete blood count; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; HbA1c, glycated hemoglobin; PT/INR, prothrombin time/international normalized ratio; TSH, thyroid-stimulating hormone.

Figure 3

Results of mixed-effect regression analysis for specific laboratory testing rates. (a) Ten-year time trends. (b) Effect sizes of patient age. (c) Effect sizes of patient sex. (d) Between-general practitioner variance in terms of the null-model intraclass correlation coefficient. Abbreviations: CBC, complete blood count; CRP, C-reactive protein; ESR, erythrocyte sedimentation rate; HbA1c, glycated hemoglobin; PT/INR, prothrombin time/international normalized ratio; TSH, thyroid-stimulating hormone.

Figure 4

General practitioner-level random effect distribution. Each single point indicates the patient age- and sex-adjusted random effect estimate for one general practitioner (GP; n = 389) for the overall laboratory testing rate. Such a random effect estimate is numerically equivalent to the estimated log OR for laboratory testing during a consultation of a given patient by the corresponding GP relative to a rate given by the fixed intercept estimate of the null model (Table 2). The difference between the x-coordinates of any two point estimates can therefore be interpreted as the log OR between the corresponding GPs for laboratory testing during a consultation of a given patient.