Prescribing antibiotics: Factors associated with C-reactive protein testing in general practice. A register-based study

The use of C-reactive protein (CRP) tests is shown to safely reduce antibiotic prescribing for 18 acute respiratory tract infections (RTIs). The aim of this study was to explore patient and clinical 19 factors associated with the use of CRP testing when prescribing antibiotics recommended for 20 RTIs. A nation-wide retrospective cross-sectional register-based study based on first redeemed 23 antibiotic prescriptions issued to adults in Danish general practice between July 2015 and June 24 2017. Only antibiotics recommended for treatment of RTIs were included in the analysis 25 (penicillin-V, amoxicillin, co-amoxicillin or roxithromycin/clarithromycin). Logistic regression 26 models were used to estimate odds ratios for patient-related and clinical factors on performing a 27 CRP test in relation to antibiotic prescribing.


48
Most acute respiratory tract infections (RTIs) are either non-severe or of viral origin and often 49 the immune system is capable of controlling the infection without antibiotics. Still, these 50 infections are frequently treated with antibiotics although it will often only add marginal benefits

52
Refraining from using antibiotics in the mildly to moderately ill patients, where modest benefit 53 can be expected, can minimise the risk of side effects for these patients and in a societal 54 perspective reduce antibiotic resistance and costs (4-6). This approach will preserve the 55 effectiveness of antibiotics and ensure relevant treatment of serious infections in the severely ill.

82
This nationwide study is a retrospective cross-sectional register-based study linking Danish 83 national registers for the adult population. Data were linked at patient-level using encrypted civil 84 registration numbers.

85
Setting 86 Health services in Denmark are tax-funded and medical expenditures partly subsidised. About 87 98% of all Danish citizens are registered with a GP. GPs are remunerated through a mixed 88 capitation and fee-for-service system with fees for a consultation and additional fees for 89 performing different services including CRP tests. The out-of-hour services (OOHS) are 90 organised by the GPs in four of the five Danish Regions (21). In the fifth region the OOHS is 91 organised by the Regional health care service.

Study population, data sources and variables for the study 93
The Danish National Prescription Registry was used to define the population. This database

101
Solely prescriptions issued in general practice or OOHS were included in this study. If a patient 102 had redeemed several antibiotic prescriptions during the project period, only information about the first prescription was used for the analyses. A 14-days antibiotic free-period was required for 104 inclusion in the study. All information was linked using encrypted unique person identification 105 numbers.

106
In order to define the diagnosis for which each prescription was issued, indication codes were 107 used. These codes represent the indication stated on the prescription by the prescriber. We 108 categorised the prescriptions according to stated indication.

109
The primary outcome measure for the study was the binary variable: whether or not a CRP test    year. If neither were available, the patient was assigned to the largest group.

137
From the Service Provider Register we retrieved speciality codes to identify GPs. We used this 138 information to take clustering at practice level into account.

Statistical analysis 140
Using descriptive statistics, we generated information on characteristics related to the

152
Characteristics of patients redeeming an antibiotic prescription 153 Table 1   1 Defined by patient being followed in practice with at least one chronic condition in the previous year.

161
Factors associated with the use of CRP 162 Table 2   for CRP testing in relation to prescriptions during April to September.

181
When restricting analyses to prescriptions with a stated RTI (model 2), we found similar trends 182 regarding the variables as for model 1 with a few exceptions. Regarding gender, the model 2 showed no differences between males and females, and for comorbidity a Charlson Index of 1 184 or above was associated with lower odds of having a CRP test performed in relation to antibiotic 185 prescription.

186
Results of the univariable models and the sensitivity analysis for prescriptions with a possible 187 RTI indication (comprising prescriptions with RTI, 'against infection' or missing indication) are 188 presented in Appendix 2. The sensitivity analysis showed no major differences from model 1.

Main findings 193
This nationwide population-based study included 984,149 individuals who had received an  for patients who are more vulnerable due to age and/or comorbidity. Furthermore, a proportion 203 of these patients may be too old or frail to be able to attend general practice and will need home 204 visits with no access to POC tests. On the other hand, these groups are also more vulnerable to 205 side-effects of antibiotics, meaning that diagnostic certainty should be prioritised. Being followed 206 in general practice for one or more chronic condition was associated with higher odds of having 207 a CRP test performed in relation to an antibiotic prescription for an RTI. In contrast, we find 208 lower odds of CRP testing with higher comorbidity score, indicating that these two measures 209 might assess different aspects. Patients with more than four consultations in practice in the 210 previous year had also higher odds for a CRP test, which may be explained by the fact that 211 these patients are well-known in practice and are more inclined to visit general practice for a 212 test.

213
CRP tests are also less used for individuals out of workforce or on disability pension, living 214 alone, immigrants and descendants of immigrants. These findings suggest that some groups of 215 patients are treated differently with no evident differences in terms of risk of complications.

Strengths and limitations of the study 224
This study is based on nationwide registers, recognised for their high validity and ability to cover 225 the entire Danish population, hereby providing a large data material.

226
However, some limitations must be considered when interpreting the results. The study used 227 the Danish National Prescription Registry which gave us access to redeemed prescriptions.

228
Prescriptions issued, but not redeemed, were not accessible. However, a previous Danish study 229 have found that primary non-adherence for antibiotics is only 6.5% (24).

230
Another Danish study found that 32% of antibiotic prescriptions had a missing indication (16).

231
The indications used in this study are the ones stated by the GPs on the prescriptions and only 232 13% were missing. We attempted to account for this by estimating two models, one with all 233 prescriptions and one with the subgroup of prescriptions with RTI as stated indication. The 234 multivariable models were adjusted for all covariables, since these variables were selected 235 based on hypothesis of influencing the main outcome. We did not take time trends into account, 236 since the study period only covered two years, and as a recent study found that frequency of 237 use of CRP tests did not change over the study period (14).

238
The database was restricted to patients who redeemed an antibiotic recommended for RTIs.

239
Hence, we were not able to assess patients for whom the GP chose to perform a CRP test but

246
The study uses temporal links between CRP tests and antibiotic prescriptions. Data do not 247 confirm whether it is the same GP who performed the test and issued the prescription. In 248 addition, we allowed a timespan of up to 7 days between CRP test and prescription. This approach was necessary since the use of CRP tests are reported on a weekly basis. However,

250
we consider this timespan clinically appropriate and would not expect it to constitute a large 251 limitation to the study.

Findings in relation to existing knowledge 253
Antibiotic prescribing can be influenced by many factors such as patients' medical history; 254 comorbidity; clinical examination; and the result of a CRP test (13, 15, 25). We found 255 differences in the use of CRP tests among different groups of patients. Odds for using CRP 256 tests were lower for the elderly and patients with comorbidities (defined by Charlson Score

268
We found that having a CRP performed in the previous year was associated with increased 269 odds of having a CRP test in relation to an antibiotic prescription. This finding is in line with 270 another study, which found that there are differences in the use of diagnostic tests between 271 practices in the range of use of diagnostic tests (30). Whether this is related to the patient or 272 more likely explained by differences in GPs' use of tests is not possible to say from the present 273 study but could be further explored in future studies.

Implications 275
This study discovered differences in the use of CRP tests among different patient groups.

276
Socially deprived patients had lower odds of having a CRP test performed in relation to an 277 antibiotic prescription for an RTI. Further studies should attempt to get a deeper knowledge of why patients are handled differently, with a special focus on socioeconomic inequality. The

279
influence of GP factors on the decision to use a CRP test was not assessed in this study, 280 however this angle could also be an important topic for future studies.

281
Furthermore, the study findings might call for clearer guidelines for GPs on how and when to 282 apply CRP tests in clinical practice. The existing evidence on when to use the CRP test could 283 be implemented more explicitly in the clinical recommendations for treatment of RTIs. In a 284 broader perspective, it must be expected that a wider selection of POC tests will be available in 285 the future.

309
Consent to participate: Not applicable.

Consent for publication 311
Not applicable.

Availability of data and materials 313
The data that support the findings of this study are available from Statistics Denmark, but 314 restrictions apply to the availability of these data, which were used under license for the current 315 study, and so are not publicly available. Data are however available from the authors upon 316 reasonable request and with permission of Statistics Denmark.