Physicians' and pharmacies' overview of patients' medication. An analysis of fidelity coefficients more

Eur J Clin Pharmacol DOI 10.1007/s00228-011-1026-3 PHARMACOEPIDEMIOLOGY AND PRESCRIPTION Physicians’ and pharmacies’ overview of patients’ medication. An analysis of fidelity coefficients Anton Pottegård & Jesper Hallas Received: 14 November 2010 / Accepted: 22 February 2011 # Springer-Verlag 2011 Abstract Background It is essential that pharmacies and prescribers have an overview of each patient’s medication in order to prevent drug interactions, unintentional co-prescribing, unnecessary polypharmacy and underprescribing. We have assessed this overview by measuring the ‘fidelity coefficient’, a measure of the extent to which a drug user has a preference for one prescriber or one pharmacy. Methods and setting Data for all prescriptions issued for the population in Southern Denmark (population 1.2 million) in 2009 was extracted from the Odense University Pharmacoepidemiological Database (OPED). Analysis of the extracted data was then limited to persons with at least ten prescriptions within the year, resulting in 8,246,064 prescriptions issued to 283,388 individuals. For each individual, we identified the most used prescriber and calculated the proportion of all prescriptions accounted for by that prescriber (FCpresc). The individual user’s most frequented pharmacy was also identified and the FCpharm calculated in a similar fashion. Results The average FCPresc and average FCPharm were 0.883 (standard deviation 0.158) and 0.927 (0.139), respectively. The estimated difference was 0.0446 (95% confidence interval 0.0439–0.0453). Among the factors associated with a high FCpresc and high FCpharm were older age, male gender and a high volume of prescriptions. The major drug classes that were most often prescribed by a non-main prescriber were beta-lactams, antidepressants and opioids. Similarly, the major drug classes associated with A. Pottegård : J. Hallas (*) Research Unit of Clinical Pharmacology, University of Southern Denmark, JB Winsløwsvej 19,2, Odense, Denmark e-mail: jhallas@health.sdu.dk use of non-main pharmacy were beta-lactams, antidepressants and inhaled beta-agonists. Conclusion Based on this analysis, both prescribers and pharmacies generally have an equal potential for maintaining an excellent overview of their patients’ medication, but the pharmacies account for a slightly higher proportion of patients. Keywords Pharmacy . Prescriber . Fidelity coefficient . Patient’s medication . Overview Introduction The risk of adverse drug reactions, polypharmacy, drug interactions and unintentional co-prescribing has become an increasingly problem as the intake of medicine also increases [1, 2]. The aim of many interventions is to decrease these adverse events, but to do so it is necessary to have an overview of the individual patient’s medicine intake. However, several studies have revealed enormous discrepancies between the records of the general practitioner (GP), hospital admission papers, pharmacy records and the patient’s own medicine cabinet [3–10]. Among elderly patients, the number of prescribing physicians is an independent risk factor for experiencing an adverse drug event [7, 11, 12]. A study by Gilchrist et al. in 1987 already revealed that up to two thirds of the medical records pertaining to a patient’s drug history obtained from the GP were inaccurate [10]. It has since been repeatedly demonstrated that GP records [3, 5–7, 9] as well as hospital records [4, 5] and even patient reporting [4–6, 8] show major discrepancies when compared to more thorough medication reviews, with up to 25% of prescribed drugs being used without the GP’s knowledge [3]. The Eur J Clin Pharmacol results of a Danish study suggest that the use of a nationwide database may prove to be the most accurate measure of actual drug use [4]. The two central players in this field are the prescriber and the pharmacy. We have attempted to assess their overview using the ‘fidelity coefficient’ (FC), a measure of which proportion of an individual patient’s medication is accounted for by their most frequently used prescriber and pharmacy. Materials and methods Materials and setting The data for this study was drawn from the Odense University Pharmacoepidemiological Database (OPED), which is a research database with full coverage of all reimbursed prescriptions in the Region of Southern Denmark (1.2 million inhabitants). The data included in each prescription record comprises the names of prescription holder, the prescriber and the pharmacy, the date of dispensing and a full account of the dispensed product, including substance, brand name, route of administration, Anatomical Therapeutic Chemical (ATC) classification code and defined daily dose (DDD) [13]. Some drugs are completely exempt from reimbursement and thus not covered by the database, including benzodiazepines, oral contraceptives, laxatives and certain antibiotics. Drugs with any degree of co-payment are covered by the database. All prescriptions redeemed by citizens of the Region of Southern Denmark during 2009 were eligible for inclusion in the analysis. Analysis The analysis was restricted to individuals who had redeemed ten or more prescriptions during 2009. For each individual, we identified the prescriber who occurred most frequently on that individual’s prescription list. We defined the prescriber fidelity coefficient, FCPresc, as the proportion of an individual’s redeemed prescriptions that were issued by the most frequent prescriber to that individual. Similarly, we defined the pharmacy fidelity coefficient, FCPharm, as the proportion of an individual’s prescriptions that were redeemed at the most frequently used pharmacy. Unless otherwise stated, the FCPharm and FCPresc are interpreted as a characteristic of a specific person; for example, when calculating the average FCPharm, we calculated the average value for FCPharm for all individual subjects in the study. The FCPresc and FCPharm are presented using standard descriptive statistics. We also explored the dependency of FCPresc and FCPharm on such variables as age, gender, number of prescriptions, whether the most frequent prescriber was a GP, whether the main pharmacy had more than one dispensing site and whether the most used pharmacy was urban. Urban pharmacies were defined as those located in the Odense or Esbjerg municipalities (186,000 and 115,000 inhabitants, respectively) or those which had the same zip-code as another pharmacy. These associations were analysed using two linear regression models, one with FCPresc as the dependant variable and one with FCPharm as the dependant variable. Data from the following individuals were excluded from this part of the analysis: (1) all individuals who had two or more pharmacies sharing the ‘preferred’ position where at least one was near a competitor and at least one was not (n=2246); (2) all individuals who showed equal preference for both a GP and a non-GP among the preferred prescribers (n=1,779). The proportion of prescriptions either issued by a nonmain prescriber or redeemed at a non-main pharmacy was categorised according to the major drug classes. We grouped the drug classes according to the third level of the ATC code (e.g. M01A = nonsteroidal anti-inflammatory drugs (NSAIDs)). Only groups with more than 50,000 prescriptions (covering 88.7% of the data) were reported. Finally, the proportion of prescriptions issued by a nonmain prescriber or redeemed at a non-main pharmacy as a function of the month was determined, which enabled the construction of a seasonality curve for FCPresc and FCPharm. Results A total of 10,067,798 prescriptions issued to 853,217 different individuals were extracted from the OPED in 2009. After restricting the data selection to individuals with ten or more prescriptions during 2009, we obtained 8,246,064 prescriptions issued to 283,388 individuals (121,734 (42.8%) men). The median age of the study cohort was 64 years (interquartile range 52–75 years). The average FCPresc and average FCPharm were 0.882 (standard deviation 0.158) and 0.927 (0.139), respectively. The average difference was 0.0446 (95% confidence interval 0.0439–0.0453). There were 116,918 persons (41.2%) with an FCPresc of 1.00 and 182,030 (64.2%) individuals with an FCPharm of 1.00. Of those, 91,665 (32.3%) had a value of 1.00 for both parameters. The FCPharm was higher than the FCPresc for 126,585 persons (44.7%), while the reverse pattern was observed for 50,640 persons (17.7%). There were 1,683 unique main prescribers and 242 unique main pharmacies. Among the variables that were found to be significantly associated with high FCPharm were an older age, male gender, high volume of prescriptions, main pharmacy Eur J Clin Pharmacol Table 1 The dependency of the pharmacy fidelity coefficient on explanatory variables Base FCPharm Agea Male gender Number of prescriptionsb Main pharmacy near competing pharmacyc Main pharmacy having more than one dispensing site Fidelity coefficient 0.838 [0.836–0.839] 0.018 [0.017–0.018] 0.011 [0.010–0.012] 0.003 [0.003–0.003] −0.053 [−0.053 to −0.052] 0.006 [0.005–0.007] .75 .8 .85 .9 .95 1 FCPharm, Pharmacy fidelity coefficient: proportion of an individual’s prescriptions that were redeemed at the most frequently used pharmacy Data are given as the FCPharm, with the 95% confidence interval (CI) given in parenthesis a b 0 20 40 60 80 100 Age Women, pharmacy Men, pharmacy Women, prescriber Men, prescriber The influence of age over FCPharm is given as the change per 10 years Fig. 1 The dependency of the fidelity coefficient on age and sex The influence of number of prescriptions over FCPharm is given as the change per 10 prescriptions The classification of ‘nearby pharmacies’ is given in the Materials and methods section c of non-main pharmacy were beta-lactams, antidepressants and adrenergics (inhalants) (Table 3). Discussion having more than one dispensing site and the use of a pharmacy with no competing pharmacies nearby (Table 1). The use of a pharmacy near a competitor was associated with a 0.053 lower FCPharm than the use of other pharmacies. When the analysis was restricted to only pharmacies near a competitor, the crude FCPharm was 0.894. The variables associated with high FCPresc was older age, male gender, high number of prescriptions and use of a GP as the main prescriber (Table 2). The dependency of FCPharm and FCPresc on age and sex is shown in Fig. 1. Figure 2 shows the seasonality of both measures. The major drug classes that were most often prescribed by a non-main prescriber were beta-lactams, antidepressants and opioids. Similarly, the major drug classes associated with use The average FCPharm and average FCPresc were 0.927 and 0.882, respectively. Thus, there is a slightly higher fidelity towards the pharmacy than towards the main prescriber. However, both have the possibility to have an excellent overview of their clients’ medication. The actual overview also depends on factors such as the structure and the interface of the IT-solutions used by the prescriber and the pharmacies and the training of the prescribers and pharmacist. Also, our analysis is based on the actual dispensing of drugs; therefore, we have no means of assessing to which Table 2 The dependency of the prescriber fidelity coefficient on explanatory variables Base FCPresc Main prescriber being a general practitioner Agea Male gender Number of prescriptionsb 0.627 0.200 0.012 0.002 0.002 [0.625–0.630] [0.198–0.202] [0.011–0.012] [0.001–0.003] [0.002–0.002] Fidelity coefficient .8 1 .85 .9 .95 1 2 3 4 5 6 7 8 9 10 11 12 FCPresc, Prescriber fidelity coefficient: proportion of an individual’s redeemed prescriptions that were issued by the most frequent prescriber to that individual Data are given as the FCPresc, with the 95% confidence interval (CI) given in parenthesis a b Month FCpharm FCpresc The influence of age over FCPharm is given as the change per 10 years The influence of number of prescriptions over FCPharm is given as the change per 10 prescriptions. Fig. 2 Season variability of the fidelity coefficient (FC). FCPresc proportion of an individual’s redeemed prescriptions that were issued by the most frequent prescriber to that individual, FCPharmproportion of an individual’s prescriptions that were redeemed at the most frequently used pharmacy Eur J Clin Pharmacol Table 3 The major drug classes to be prescribed by other than main prescriber and redeemed at other than main pharmacies ATC ATC-text Total number of prescriptions 545,541 531,388 449,645 392,559 355,135 Prescriptions redeemed at a pharmacy other than the main pharmacya 6.42 (35,021) 8.68 (46,126) 8.87 (39,888) 6.31 (25,851) 4.82 (18,923) 7.96 (28,282) 8.27 (28,381) 8.86 (29,016) 6.83 (21,839) 6.32 (17,187) 6.39 (16,031) 6.25 (15,500) 13.18 (30,586) 6.01 (12,846) 31.28 (65,710) 4.93 (10,241) 10.69 (21,322) 5.45 (9,271) 8.04 (12,654) 15.68 (22,693) 6.85 (7,848) 11.68 (11,721) 6.39 (5,611) 43.24 (36,562) 10.7 (28,773) 21.03 (17,189) 5.45 (4,376) 16.56 (12,828) 5.57 (4,107) 21.84 (15,818) 6.66 (4,585) 8.08 (5,512) 5.75 (3,711) 7.22 (4,319) 12.26 (7,024) 11.40 (5,997) 23.26 (12,202) 9.87 (5,169) Prescriptions issued by a prescriber other than the main prescribera 3.26 (17,784) 5.57 (29,622) 5.42 (24,353) 3.78 (15,503) 3.51 (13,774) 7.03 (24,957) 4.98 (17,077) 6.55 (21,465) 3.68 (11,775) 4.00 (10,887) 3.90 (9,795) 4.33 (10,735) 4.42 (10,265) 2.54 (5,423) 15.29 (32,127) 3.63 (7,547) 5.25 (10,467) 2.68 (4,568) 6.46 (10,162) 6.23 (9,024) 4.38 (5,017) 6.67 (6,688) 4.84 (4,249) 6.56 (5,547) 8.02 (6,560) 9.17 (7,495) 4.35 (3,489) 7.95 (6,156) 4.43 (3,262) 10.33 (7,483) 3.16 (2,173) 11.15 (7,612) 2.75 (1,774) 2.97 (1,776) 3.81 (2,184) 5.15 (2,708) 14.43 (7,571) 3.74 (1,957) B01A Antithrombotic agents N06A Antidepressants N02A Opioids N02B Non-opioid analgesics and antipyretics R03A Adrenergics, inhalants C10A Cholesterol and triglyceride reducers 409,749 A02B Drugs for peptic ulcer and gastro343,252 oesophageal reflux disease M01A Anti-inflammatory and antirheumatic 327,479 products, non-steroids C07A Beta blocking agents 319,839 C09A ACE inhibitors, plain C08C Selective calcium channel blockers with mainly vascular effect A10B Oral blood glucose lowering drugs N05A Antipsychotics C03C J01C High-ceiling diuretics 272,042 250,901 247,964 232,120 213,882 207,753 199,453 170,159 157,373 144,746 114,579 100,345 87,781 84,547 81,832 81,743 80,269 77,459 73,710 72,413 68,863 68,245 64,522 57,270 52,594 52,457 52,377 Beta-lactam antibacterials, penicillins 210,077 C03A Low-ceiling diuretics, thiazides N03A Antiepileptics A12B Potassium R03B Other drugs for obstructive airway diseases, inhalants A10A Insulins and analogues C09C Angiotensin II antagonists, plain G03C Oestrogens H03A Thyroid preparations S01E Antiglaucoma preparations and miotics R06A Antihistamines for systemic use D07A Topical corticosteroidss, plain C09D Angiotensin II antagonists, combinations H02A Corticosteroids for systemic use, plain C09B ACE inhibitors, combinations R01A Decongestants and other nasal preparations for topical use M05B Drugs affecting bone structure and mineralisation N02C Antimigraine preparations C01A Cardiac glycosides N04B Dopaminergic agents G04B Other urologicals, incl. antispasmodics J01F Macrolides, lincosamides and streptogramins G04C Drugs used in benign prostatic hypertrophy C01D Vasodilators used in cardiac diseases 59,812 ATC, Anatomical Therapeutic Chemical (ATC) classification code The data have been sorted by number of prescriptions. Only groups with more than 50,000 prescriptions are included (covering 88.7% of our data) a Data on prescriptions are given as the percentage with the number of prescriptions given in parenthesis Eur J Clin Pharmacol extent the main prescriber is made aware of the prescriptions issued by specialists or other doctors to his patients. The primary strength of the study is the high internal validity due to a high quality of the prescription data [13]. In addition, there is little selection bias since all residents of Region of Southern Denmark were included in the analysis. The primary weakness of the study is that the FCPharm and FCPresc are to a large extent determined by the underlying healthcare structure. Our results may thus not necessarily be equally applicable to another setting. There are several factors in our setting that would favour a high FCPharm over the FCPresc. First, pharmacies in Denmark are large units, often covering a substantial area, especially in comparison with the average pharmacy found, for example, in southern Europe. There are 56 community pharmacies in the region covered in this analysis, corresponding to a density of one pharmacy per 21,400 citizens. It is noteworthy, however, that the FCPharm only shows a minor dependency on having multiple pharmacies nearby (Table 1). Furthermore, many doctors are specialists and thus only responsible for prescribing a minor part of a patient’s total medication. Other factors favour the FCPresc over the FCPharm. Pharmacies are completely liberalised in Denmark and patients are therefore free to choose between pharmacies. In contrast, each citizen is assigned a regular GP who serves as a gate keeper, which means that all medical contacts, excluding emergencies, should go through the assigned GP. Although it is possible to change GP, this happens relatively rarely. There is also a tendency in Denmark among GPs to form larger units consisting of several GPs under the same roof and using the same prescriber identifier. As such, the single prescriber ID in our analysis can cover more than one individual prescriber. As these prescribers can see each other's prescribing to the individual patient within the group practice , they have the opportunity to avoid such problems that relate to multiple prescribers. Also, repeat prescriptions were registered as multiple single prescriptions in our analyses, even though they only represent a single prescription decision. By definition, repeat prescriptions are issued by the same prescriber, but not necessarily redeemed at the same pharmacy, which is also a factor that would favour a high FCPresc relative to the FCPharm. Finally, GPs frequently take over the prescribing of specialised drug regimes as soon as the medication is stable. Consequently, the fidelity coefficient is highly dependent on the healthcare structure. Most of the factors in our setting point towards a higher FCPresc than FCpharm. It is therefore surprising that the results of our analysis reveal a FCPharm greater than the FCPresc. Our analysis in Table 3 shows that antibiotics account for most of the infidel prescriptions, a result which is hardly surprise. It is more interesting to note that the groups of ‘antidepressants’, ‘antipsychotics’ and ‘antithrombotic agents’ were so highly represented. These three groups are known to often represent long-term treatments and also show a wide range of possibly dangerous drug–drug interactions, especially with respect to antithrombotic agents [14]. Combining the numbers for these three groups revealed that while 111,733 of these prescriptions were prescribed by others than the most used prescriber, only 57,671 were redeemed at a pharmacy other than the most frequently used pharmacy. While both numbers are higher than desired, this results emphasises the central role of the pharmacy in identifying and preventing drug–drug interactions. The importance of the fidelity coefficient for monitoring medication profiles with the aim of avoiding doubling of prescriptions or adverse drug interactions is most obvious in a setting where data on the medication of an individual are not readily available for the healthcare practitioner. This is still the case in most countries. In Denmark, each redemption of a prescription is registered, but no complete list of ‘current treatment’ is produced for routine care. This will probably change in the coming years ash new IT-solutions appear [15]. Several questions arise from this study. First, it would be interesting to explore how the ‘fidelity coefficient’ differs across different populations and different healthcare models. It might even be possible, through subsequent studies, to link the ‘fidelity coefficient’ to other parameters, such as adverse drug event rates, on a population scale. Lastly, the ‘fidelity coefficient’ could be used as a tool to refine future population-based analyses, for example, by having a high fidelity as an exclusion or inclusion criteria. References 1. Madsen HK, Hallas J (2009) Udvikling I lægemiddelforbruget I Danmark (in Danish). Ugeskr Laeger 171(19):775–777 2. Department of Health (2004). Prescriptions dispensed in the community: statistics for 1993–2003. London, Department of Health. Available at: http://www.dh.gov.uk/prod_consum_dh/ groups/dh_digitalassets/@dh/@en/documents/digitalasset/ dh_4094068.pdf 3. Barat I, Andreasen F, Damsgaard EMS (2000) The consumption of drugs by 75-year-old individuals living in their own homes. Eur J Clin Pharmacol 56:501–509 4. Glintborg B, Poulsen HE, Dalhof KP (2007) The use of nationwide on-line prescription records improves the drug history in hospitalized patients. Br J Pharmacol 65(2):265–269 5. Foss S, Schmidt JR, Andersen T, Rasmussen JJ, Damsgaard J, Schaefer K, Munck LK (2004) Congruence on medication between patients and physicians involved in patient course. Eur J Clin Pharmacol 59:841–847 6. Glintborg B, Andersen SK, Poulsen HE (2009) Prescription data improve the medication history in primary care. Qual Saf Health Care 19(3):164–168 7. Torrible SJ, Hogan DB (1997) Medication use and rural seniors who really knows that whey are taking? Can Fam Physician 43:893–898 Eur J Clin Pharmacol 8. Warholak TL, McCulloch M, Baumgart A, Smith M, Fink W, Fritz W (2009) An exploratory comparison of medication lists at hospital admission with administrative database records. J Manag Care Pharm 15(9):751–758 9. Tulner LR, Kuper IMJA, Frankfort SV, van Campen JPCM, Koks CHW, Brandjes DPM, Beijnen JH (2009) Discrepancies in reported drug use in geriatric outpatients: relevance to adverse events and drug-drug interactions. Am J Geriatr Pharmacother 7 (2):93–104 10. Gilchrist WJ, Lee YC, Tam HC, MacDonald JB, Williams BO (1987) Prospective study of drug reporting by general practitioners for an elderly population referred to a geriatric service. Br Med J 294:289–290 11. Tamblyn RM, McLeod PJ, Abrahamowicz M, Laprise R (1996) Do to many cooks spoil the broth? Multiple physician involvement in medical management of elderly patients and potentially inappropriate drug combinations. Can Med Assoc J 154(8):1177– 1184 Green JL, Hawley JN, Rask KJ (2007) Is the number of prescribing physicians an independent risk factor for adverse drug events in an elderly outpatient population? Am J Geriatr Pharmacother 5(1):31–39 Gaist D, Sørensen HT, Hallas J (1997) The Danish prescription registries. Dan Med Bull 44(4):445–448 Hallas J, Dall M, Andries A, Andersen BS, Aalykke C, Hansen JM, Andersen M, Lassen AT (2006) Use of single and combined antithrombotic therapy and risk of serious upper gastrointestinal bleeding: population based case-control study. Br Med J 333 (7571):726–730 Bereznicki BJ, Peterson GM, Jackson SL, Walters EH, Fitzmaurice KD, Gee PR (2008) Data-mining of medication records to improve asthma management. Med J Aust 189(1):21–25 12. 13. 14. 15.