Skip to main content
PLOS Medicine logoLink to PLOS Medicine
. 2020 Jun 26;17(6):e1003151. doi: 10.1371/journal.pmed.1003151

Impact of physicians’ participation in non-interventional post-marketing studies on their prescription habits: A retrospective 2-armed cohort study in Germany

Cora Koch 1,2,*, Jörn Schleeff 3, Franka Techen 3, Daniel Wollschläger 4, Gisela Schott 5, Ralf Kölbel 6, Klaus Lieb 2
Editor: Sanjay Basu7
PMCID: PMC7319278  PMID: 32589633

Abstract

Background

Non-interventional post-marketing studies (NIPMSs) sponsored by pharmaceutical companies are controversial because, while they are theoretically useful instruments for pharmacovigilance, some authors have hypothesized that they are merely marketing instruments used to influence physicians’ prescription behavior. So far, it has not been shown, to our knowledge, whether NIPMSs actually do have an influence on prescription behavior. The objective of this study was therefore to investigate whether physicians’ participation in NIPMSs initiated by pharmaceutical companies has an impact on their prescription behavior. In addition, we wanted to analyze whether specific characteristics of NIPMSs have a differing impact on prescription behavior.

Methods and findings

In a retrospective 2-armed cohort study, the prescription behavior of 6,996 German physicians, of which 2,354 had participated in at least 1 of 24 NIPMSs and 4,642 were controls, was analyzed. Data were acquired between 6 October 2016 and 8 June 2018. Controls were matched by overall prescription volume and number of prescriptions of the drug studied in the NIPMS in the year prior to the NIPMS. Primary outcome was the relative rate of prescriptions of the drug studied in the NIPMS by participating physicians compared to controls during the NIPMS and the following year. Secondary outcomes were the proportion of prescriptions of the studied drug compared to alternative drugs used for the same indication, the revenue generated by these prescriptions, and the association between the marketing characteristics of the NIPMS and prescription habits. Of the 24 NIPMSs, the 2 largest drug groups studied were antineoplastic and immunomodulatory agents (7/24, 29.2%) and agents for the nervous system (4/24, 16.7%). Physicians participating in an NIPMS prescribed more of the studied drug during and in the year after the NIPMS, at a relative rate of 1.08 (95% CI 1.07–1.10; p < 0.001) and 1.07 (95% CI 1.05–1.09); p < 0.001), respectively. Participating physicians were more likely than controls to prescribe one of the studied drugs rather than alternative drugs used for the same indication (odds ratio 1.04; 95% CI 1.03–1.05). None of the marketing characteristics studied were significantly associated with prescription practices. The main limitation was the difficulty in controlling for confounders due to privacy laws, with a resulting lack of information regarding the included physicians, which was mainly addressed by the matching process.

Conclusions

Physicians participating in NIPMSs prescribe more of the investigated drug than matching controls. This result calls the alleged non-interventional character of NIPMSs into question and should lead to stricter regulation of NIPMSs.


Cora Koch and colleagues reveal the influence of pharmaceutical marketing campaigns on doctors' prescribing habits.

Author summary

Why was this study done?

  • After drugs are authorized, non-interventional post-marketing studies (NIPMSs) are initiated to study rare side effects or other aspects of the drug that may have been missed during the authorization trials.

  • Previous studies have shown that NIPMSs, while conducted regularly, often lack scientific rigor, rarely lead to relevant results, and are rarely published.

  • Some authors have therefore hypothesized that NIPMSs primarily serve marketing purposes for pharmaceutical manufacturers by familiarizing physicians with new drugs.

  • So far, it is unclear whether NIPMSs actually do have an impact on physicians’ prescribing behavior.

What did the researchers do and find?

  • We conducted a study in Germany comparing the prescription behavior of 2,354 physicians who had participated in at least 1 of 24 NIPMSs and 4,642 comparable physicians who had not participated in such studies.

  • We found that physicians participating in an NIPMS prescribed 6%–8% more of the drug studied in the NIPMS than comparable physicians during the NIPMS and the year after.

  • We also looked at certain characteristics of the NIPMSs to see whether they predicted the impact on prescription behavior, but found no characteristics that were associated with the impact.

What do these findings mean?

  • NIPMSs seem to have an impact on physicians’ prescribing behavior despite their “non-interventional” nature.

  • Up to this point, NIPMSs have been only very loosely regulated because it was assumed that they have a low potential to cause harm. However, with the possibility that physicians prescribe differently due to their participation in an NIPMS, which may or may be detrimental to patients, only NIPMSs that are designed to collect essential data should be permitted.

Introduction

Non-interventional post-marketing studies (NIPMSs) funded by the pharmaceutical industry have been the subject of controversial debate. In principle, their purpose is to provide data on the real-world safety and/or effectiveness of recently authorized drugs by studying them with a larger and less highly selected patient population than is usual in authorization trials [1,2]. Some NIPMSs are imposed by regulatory agencies, for example to assess a safety risk of a medicinal product or to evaluate the effectiveness of risk management measures [3]. However, some authors have hypothesized that many NIPMSs primarily serve marketing purposes for pharmaceutical manufacturers by familiarizing physicians with a newly authorized drug as well as offering an incentive to prescribe the drug [1,47].

A recent study of German NIPMSs by Spelsberg et al. showed that they rarely serve to improve drug safety because their sample sizes are usually too small to allow for the detection of rare adverse events [8]. Other studies have also pointed in this direction by showing that NIPMSs usually lack scientific quality [4,5,9]. In addition, Spelsberg and colleagues raised the concern that strict confidentiality clauses in combination with the high remuneration for participation in NIPMSs could actually serve to discourage physicians from reporting adverse events [8]. NIPMSs are also published extremely rarely, even though they are conducted regularly, which makes it even less likely that they will lead to safer prescription practices, even if relevant data are generated [8,1012]. Another concern is the effect the participation in NIPMSs could have on physician prescription practices. Physicians’ participation in an NIPMS usually consists of enrolling patients who are prescribed a certain drug in an NIPMS and gathering data on the enrolled patients regarding parameters such as adverse events or efficacy. For the enrollment of each patient, physicians are remunerated with a certain fee. By remunerating physicians for the inclusion of a patient who is prescribed a certain drug, NIPMSs offer an incentive for the prescription of this drug, possibly affecting prescription practices.

Although several authors have argued that NIPMSs mainly serve the marketing purposes of the pharmaceutical industry, it has not yet been demonstrated, to our knowledge, that the prescription behavior of physicians indeed changes during or after participation in an NIPMS; such an effect has only been demonstrated for interventional studies [1315]. However, if this were the case, it would be an important reason to increase regulation of such studies. Currently, NIPMSs do not need to be registered in the US; in Germany, they need to be registered, but they do not need to be authorized by a higher federal authority, as interventional clinical trials need to be [16]. Concerns that such studies may not only be less useful than they are made out to be, but may in addition have deleterious effects on physicians’ reporting of adverse events and their prescription practices should lead to stricter scrutiny of such studies before they are initiated.

The objective of the current study was thus primarily to investigate whether participation in NIPMSs impacts the participating physicians’ prescription practices. In addition, we wanted to analyze whether such a possible change results in more expensive prescriptions by leading to a shift in prescriptions toward more expensive drugs when there are less expensive alternative drugs used for the same indication. Lastly, we wanted to investigate whether certain characteristics of NIPMSs are useful to predict the impact on the participating physicians’ prescription behavior.

Methods

Ethics review

In a previous study [9], we gathered data on NIPMSs at the National Association of Statutory Health Insurance Funds (GKV-Spitzenverband), of which some data were used in the current study. The local ethics committee of the Landesärztekammer Rheinland-Pfalz decided that it was not necessary to conduct an ethics review for this previous study. To gain access to the prescription data of physicians participating in NIPMSs as well as controls, we submitted a request to the German Federal Ministry of Health that the National Association of Statutory Health Insurance Funds as well as a specific German statutory health provider (the Innungskrankenkasse) be allowed to provide us with the respective data. This request was the basis for the planning of the study (see S1 File). After this request was granted, we again consulted with the local ethics committee, which again decided that it was not necessary to conduct an ethics review for the current study.

Study design

In a retrospective 2-armed cohort study, we compared the prescription practices of physicians who had participated in an NIPMS with those of matched controls who had not participated in an NIPMS. After identifying eligible NIPMSs, participating physicians were identified and matched 1:2 to control physicians, resulting in a “matching group” (see Table 1 for eligibility criteria). Prescription data regarding overall prescription volume (i.e., number of packages of all drugs prescribed by a physician) as well as prescriptions of the drug studied in the NIPMS and alternative drugs were acquired for the year before the NIPMS (t0), during the NIPMS (t1), and the following year (t2) (see S1 Appendix for definition of “alternative drug”). Differences in prescription volume during and after the study regarding the studied drugs as well as alternative drugs were used to assess the impact of participation in an NIPMS on prescription practices. In addition, we collected data on the following NIPMS characteristics, which could potentially be indicators of NIPMSs being conducted for marketing purposes: inappropriate remuneration, the medication having been on the market too long, low scientific quality, low formal quality, negligible effort required of physician, missing report regarding the results of the NIPMS, and presence of a secrecy clause in the contract for participating physicians (see S4 Appendix and [9] for further details). These indicators were summarized in a “marketing score” as described in our previous publication, where a higher score indicates a higher likelihood of the NIPMS having been initiated for marketing purposes (see S4 Appendix and [9]). Associations between these characteristics and prescription volume were used to assess whether NIPMSs found to be more likely to have been initiated for marketing reasons had a higher impact on prescription practices.

Table 1. Inclusion criteria for NIPMSs.

Inclusion criteria for NIPMSs Reason
Began after 31 December 2012 To allow for analysis 1 year prior to the beginning of the NIPMS (t0), as prescription data at the National Association of Statutory Health Insurance Funds were only available beginning in January 2012
Ended after 31 December 2013 Change in regulations affecting all studies ending after this date that allowed for tracking prescription practices
Ended before 1 July 2015 To allow for a follow up of 1 year (t2) immediately after the end of the NIPMS, while ending before the commencement of our data collection
Observed medicinal product or drug is covered by statutory health insurance Otherwise, notification to the National Association of Statutory Health Insurance Funds is not required regarding participating physicians
Observed medicinal product or drug is subject to a prescription by a physician and dispensed by a pharmacy Otherwise, no prescription data are available
Study is prospective Retrospective studies are not expected to affect prescription practices
Observed medicinal product or drug was approved at least 6 months before the beginning of the NIPMS Otherwise, prescription data before the study would not represent the physicians’ prescription habits accurately
Study was conducted among physicians in private practice (rather than physicians employed in hospitals) Prescription data are only available for physicians in private practice

NIPMS, non-interventional post-marketing study.

NIPMSs

We considered NIPMSs to be eligible for our study when certain criteria were met that allowed for an assessment of the prescription practices of the physicians between 1 year before and 1 year after the NIPMS. Criteria that enabled these analyses and reasons are given in Table 1.

All physicians participating in one of the eligible NIPMSs who could be identified using the data at the National Association of Statutory Health Insurance Funds and our validation process (see below) and had prescribed a minimum amount overall (to ensure they were still practicing) as well as a minimum amount of the drug studied in the NIPMS were included in the study. We matched 2 controls to each of these physicians using the number of overall prescriptions (in packages) as well as the number of defined daily doses (DDDs) of the studied drug and alternative drugs in the year before the NIPMS began (t0). Due to data privacy laws in Germany, other factors such as physician age, gender, location, and specialization could not be considered for matching. Physicians participating in several different NIPMSs were matched to different controls for each NIPMS (see S2 Appendix for exact process). Two controls were chosen for each case, as opposed to more or fewer, to balance the increase in statistical power through the number of controls with the matching quality given the limited pool of good matches for each case [17,18].

Data sources and setting

Data regarding the NIPMSs as well as participating physicians were identified from notifications submitted to the National Association of Statutory Health Insurance Funds regarding NIPMSs. In Germany, companies planning such a study need to notify 3 different authorities before its initiation, among others the National Association of Statutory Health Insurance Funds. The notification needs to include location, beginning and end dates, and the objective of the study as well as a list of participating physicians (by name and by LANR, a unique ID number permanently assigned to each physician) and a study plan. In addition, for studies observing a drug or medicinal product that is covered by statutory health insurance, companies need to provide information regarding type and amount of remuneration received by the participating physicians as well as a sample contract between the company and the physicians. The amount of remuneration needs to be justified by the effort required by the physician, and this justification must be described by the company in the notification. Data regarding the characteristics of the studies as well as the identifying data on physicians were gathered between 6 October 2016 and 9 January 2018. Because notifications were often incomplete or faulty regarding the identifying information on physicians, we conducted a validation by comparing the acquired data with a directory at the Innungskrankenkasse containing all practicing physicians in Germany. This validation process took place between 22 and 26 January 2018. See S3 Appendix for a precise description of this process.

Prescription data were acquired using data from the GKV-Arzneimittelschnellinformation (GAmSi) project, which consists of data reported to the National Association of Statutory Health Insurance Funds by pharmacies regarding filled prescriptions for patients with statutory health insurance (in 2018, 87% of German citizens were insured by one of the statutory health insurance providers [19]). For each physician, we acquired data regarding the drug studied in the NIPMS they participated in or were matched to as controls, as well as alternative drugs and overall number of prescriptions before, during, and after the NIPMS. Prescription data were enriched with the official version of the German Anatomical Therapeutic Chemical (ATC) classification and DDD published by the Wissenschaftliches Institut der AOK (WIdO, Version 49, 201803) [20]. Matching and acquiring the prescription data took place between 27 January and 8 June 2018.

Outcome measures

The primary outcome was the relative rate of prescriptions of the drug studied in the NIPMS by participating physicians compared to their respective controls during the NIPMS and the year after.

Secondary outcomes were the proportion of prescriptions of the drug under study compared to alternative drugs, as well as the revenue generated by these prescriptions. In addition, the association between the marketing characteristics of the NIPMS and prescription habits was a secondary outcome.

Bias

We used directed acyclic graph modeling to identify the minimal sufficient adjustment set of covariates in the regression model to achieve unbiased estimation of the putative causal effect of participation in an NIPMS on prescription volume at t1 [21]. If confounders have a direct effect on the exposure, but only an indirect effect on the outcome via prescription volume at t0, then matching on prescription volume at t0 is sufficient for unbiased estimation (see S1 Fig for the directed acyclic graph model). Other confounders could only introduce bias through a direct effect on prescription volume at t1. Such a confounder would have to act differently on prescription volume at t1 compared to its effect on prescription volume at t0. This would seem to be implausible for candidate confounders such as age, gender, and specialization of physicians that could not be considered due to data privacy laws in Germany.

To account for confounding by physician prescription habits (in general as well as of the drug studied in the NIPMS) before entering into the NIPMS, we used prescription metrics as matching parameters. We assumed prior prescription behavior for the studied drug to be a good indicator of the previous interest of the physicians in the studied drug. In addition, in the statistical analysis, results were adjusted for prescription practices before entering the NIPMS as well as for total prescription volume during and after the NIPMS.

Sample size

Because there are no prior studies to our knowledge of changes in prescription practices due to NIPMSs, we had no plausible effect size estimate, and were therefore not able to calculate a target sample size. We thus aimed to include all studies within the time frame with available prescription data that matched our inclusion criteria.

Statistical methods

Matching groups without a participating physician and without at least 1 control were excluded from data analysis. Mean number of package prescriptions and DDDs were first calculated within each NIPMS, and then averaged across NIPMSs, weighted by the number of contributing matching groups within each NIPMS. Matching-group-wise differences were first averaged within each NIPMS, and then averaged across NIPMSs, weighted by the number of contributing matching groups within each NIPMS. The weighting scheme was used to ensure that more reliable estimates based on NIPMSs with more matching groups had more influence than more uncertain estimates based on NIPMSs with fewer matching groups.

Conditional Poisson regression was used to assess the relative rate of prescriptions for the studied drug in participating physicians versus their respective controls [22]. Coefficients for matching groups were treated as nuisance variables and were eliminated. To account for possible overdispersion, a quasi-Poisson approach was chosen. The log number of days of the NIPMS was used as offset. Regression models for t0 were adjusted for the total number of prescriptions during t0. Regression models for t1 and t2 were adjusted for the number of prescriptions for the studied drug during t0, and for the total number of prescriptions during t1 and t2, respectively.

Mixed binomial logistic regression was used to assess a shift in the proportion of prescriptions of studied drugs relative to alternative drugs used for the same indication in participating physicians versus controls. The model included a random intercept effect for matching group and allowed for 0 inflation since a relevant number of matching groups had 0 prescriptions for the drug under study. This analysis was only carried out for t1, in the interest of parsimony.

Logistic regression using generalized estimating equations was used to assess the association of marketing indicators with the probability that a participating physician prescribed more of the drug under study than the average of the corresponding controls. Clusters were defined by NIPMS, with the assumption of compound symmetry for the correlation structure.

p-Values less than 0.05 were considered statistically significant. Data were analyzed using the R environment for statistical computing version 3.5.2 with packages gnm, brms, and geepack [2326].

Sensitivity analyses

In general, we assumed that physicians had participated in an NIPMS for the entirety of the study, even though it is likely that some physicians were recruited for participation after the NIPMS had begun or stopped participating before it was officially terminated. For a subset of physicians, a more precise time frame of participation could be inferred; this was the case when NIPMS sponsors regularly reported on the participating physicians, and it was possible to determine when a specific physician first entered the study and when they stopped participating. Sensitivity analyses were conducted for this subset of physicians. In addition, sensitivity analyses were conducted that included only the drug manufactured by the sponsor of the study, as generics were available for some of the drugs studied in the NIPMSs.

Results

Study population

Of a total of 95 registered NIPMSs that began and ended in the predefined time frame, 24 matched our inclusion criteria, and 2,354 physicians that had participated in those NIPMSs could be analyzed (See S1 Fig and S2 Table for information on exclusion of physicians and NIPMSs, respectively). The mean duration of NIPMSs was 500 days (SD 181 days). The mean marketing score was 2.4 (SD 1.5; range 0–5, maximum possible value 7.5) (see Tables 2 and S2 for characteristics of individual NIPMSs). For 1,286 physicians participating in 9 NIPMSs, we could define a more specific time frame of participation in the NIPMS to conduct sensitivity analyses.

Table 2. List of NIPMSs and selected characteristics.

NIPMS Studied substance(s) Number of participating physicians Start date Duration (days) Marketing score
1 Fluocinolone acetonide 1 5 Nov 2013 529 1.5
2 Mometasone 2 3 Feb 2014 210 0.5
3 Paclitaxel 4 1 Mar 2013 731 5
4 Telaprevir 5 6 May 2013 756 1.5
5 Sorafenib 7 16 Jul 2013 708 2.5
6 Filgrastim/pegfilgrastim 7 24 Jan 2013 397 3
7 Infliximab/golimumab 16 18 Mar 2013 823 3.5
8 Tapentadol 16 1 Apr 2013 609 3.5
9 Darbepoetin alfa 19 15 Jan 2013 775 5
10 Docetaxel 21 1 Mar 2013 731 5
11 Denosumab 40 31 Jan 2013 415 2.5
12 Infliximab 43 23 Jan 2013 555 2.5
13 Ciclosporin 46 1 Jan 2014 546 2
14 Iron (III) isomaltoside 49 1 May 2013 396 4.5
15 Rasagiline 65 27 Jan 2014 339 0
16 Rivastigmine 66 15 Apr 2013 657 1.5
17 Fluorouracil and salicylic acid 117 15 Jan 2014 410 1
18 Propiverine 149 31 Jul 2014 335 0.5
19 Ingenol mebutate 171 15 Jul 2013 351 0
20 Agomelatine 219 1 Mar 2014 245 2.5
21 Testosterone 220 6 Oct 2014 268 3.5
22 Timolol and bimatoprost 281 25 Nov 2013 402 3.5
23 Ivabradine 312 17 Mar 2014 458 2.5
24 Olodaterol/tiotropium bromide 478 1 Jun 2014 365 0.5

NIPMSs for which a more precise time frame of participation could be defined and that were therefore used in sensitivity analyses (see Methods) are marked in gray.

NIPMS, non-interventional post-marketing study.

Matching quality

There was an average relative difference of 0.2% for overall number of prescribed packages and −0.83% for overall DDDs between cases and controls during time period t0. There was an average relative difference of 8.29% for number of packages and 10.13% for number of DDDs prescribed of the studied drug between cases and controls during time period t0.

Primary outcome

Participating physicians showed consistently higher absolute prescription volumes of the studied drug compared with controls, with the gap widening during t1 and narrowing during t2 (see Table 3). Relative to physicians not participating in an NIPMS, physicians participating in an NIPMS had a 7%–8% higher prescription rate of the studied drug during the NIPMS and a 6%–7% higher prescription rate during the year after the NIPMS had finished, when accounting for their overall prescription volume during the respective time frame as well as the number of prescriptions of the studied drug before the start of the NIPMS (see Table 4).

Table 3. Mean number of prescriptions of the studied drug for controls and participating physicians during the mentioned time frame, weighted by the number of matching groups in the NIPMS.

Time frame Mean duration (days)  Number of packages DDD
Control NIPMS Control NIPMS
t0 365 102.4 103.4 7,705 7,753
t1 500 109.5 120.2 8,064 8,700
t2 365 102.1 110.3 7,736 8,314

Note that t1 is not the same duration as t0 and t2; number of prescriptions may be compared between groups during the same time frame, but not between time frames. Control = matched physicians, n = 4,642; NIPMS = physicians participating in an NIPMS, n = 2,354.

DDD, defined daily dose; NIPMS, non-interventional post-marketing study.

Table 4. Estimated relative prescription rate (Rpr) of the studied drug for participating physicians versus controls.

 Time frame Number of packages DDD*
Rpr** (95% CI) p-Value Rpr (95% CI) p-Value
t0 1.04 (1.03–1.05) <0.001 1.04 (1.03–1.04) <0.001
t1 1.08 (1.07–1.10) <0.001 1.07 (1.06–1.09) <0.001
t2 1.07 (1.05–1.09) <0.001 1.06 (1.04–1.08) <0.001

Model for t0 adjusted for overall prescription rate; models for t1 and t2 adjusted for overall prescription rate and prescription rate of studied drug at t0.

*Defined daily dose (DDD) of the drug studied in the non-interventional post-marketing study.

**Relative rate; n = 2,354 groups.

Sensitivity analyses of the relative prescription rate considering only studied drugs manufactured by the sponsor (i.e., excluding generics) showed similar results, as did analyses using only data from physicians for whom a more precise time frame of participation in the NIPMS could be determined. However, in this smaller group, the difference was not statistically significant for the time period after the NIPMS (t2) regarding DDD (see S3S5 Tables for exact data).

Secondary outcomes

Shift in proportion of studied drugs and financial effect

The odds of a participating physician prescribing a drug studied in an NIPMS rather than an alternative drug used for the same indication during the time frame of the NIPMS (t1) was slightly higher than for the controls (odds ratio 1.04; 95% CI 1.03–1.05; p < 0.001). The mean revenue generated from prescriptions of NIPMS and alternative drugs during this time frame, i.e., the amount that needed to be reimbursed by statutory health insurance providers for prescriptions written by the physicians during t1, was also higher for participating physicians, with a mean revenue of 226,713€ (SD 989,087; median 32,036€) versus 153,013€ (SD 615,225; median 25,521€) for controls. However, this difference was not statistically significant, with an odds ratio for participating physicians to generate a higher revenue than controls of 1.03 (95% CI 0.98–1.08; p = 0.18).

Marketing indicators

NIPMSs scored a mean of 2.42 (SD 1.54) on the marketing score. See Table 5 for details on how many NIPMSs fulfilled each of the marketing indicators. None of the marketing indicators were significantly associated with prescription volume of the studied drug by physicians participating in an NIPMS compared to controls. While a negligible effort required for the participating physicians was associated with a lower impact on prescription volume in the univariate analysis, this difference was not confirmed by multivariate analysis. See Tables 6 and 7 for results of univariate and multivariate regression analyses, respectively.

Table 5. Number of NIPMSs fulfilling each of the marketing indicators.
Marketing indicator Number (%) of NIPMSs*
Remuneration was inappropriate or not clearly warranted 10 (41.7%)**
Drug has been on the market for too long 8 (33.3%)
Low scientific quality 10 (41.7%)
Low formal quality 5 (20.8%)
Negligible effort required by physician 1 (4.2%)
Required report missing 7 (29.1%)
Contract contains a secrecy clause 13 (54.2%)

Marketing indicators described in [9] and S4 Appendix.

*NIPMSs fulfilling the characteristics; n = 24.

**Not appropriate, 4 (16.7%); unclear whether appropriate, 6 (25.0%).

NIPMS, non-interventional post-marketing study.

Table 6. Results of univariate regression analysis regarding the relationship between marketing characteristics and probability that participating physicians prescribe more of the studied drug than controls.
Marketing characteristic Odds ratio (95% CI)
Remuneration is inappropriate or not clearly warranted 1.01 (0.79–1.30)
Drug has been on the market for too long 0.86 (0.62–1.20)
Low scientific quality 1.03 (0.80–1.33)
Low formal quality 0.91 (0.66–1.25)
Negligible effort required by physician 0.77 (0.68–0.88)
Required report missing 0.98 (0.67–1.42)
Contract contains a secrecy clause 1.15 (0.90–1.47)
Marketing score 1.00 (0.91–1.08)

Marketing indicators described in [9] and S4 Appendix. Significant difference in bold.

Table 7. Results of multivariate regression analysis regarding the relationship between marketing characteristics and probability that participating physicians prescribe more of the studied drug than controls.
Marketing characteristic Odds ratio (95% CI)
Remuneration is inappropriate or not clearly warranted 1.19 (0.65–2.17)
Drug has been on the market for too long 0.80 (0.52–1.25)
Low scientific quality 1.03 (0.82–1.28)
Low formal quality 1.06 (0.84–1.34)
Negligible effort required by physician 0.71 (0.50–1.01)
Required report missing 0.93 (0.49–1.77)
Contract contains a secrecy clause 1.04 (0.76–1.41)

Marketing indicators described in [9] and S4 Appendix.

Discussion

This study is the first to our knowledge to show in a quasi-experimental design that physicians participating in an NIPMS show changed prescription rates in favor of the investigated drug. Physicians participating in an NIPMS had a meaningfully higher rate of prescription of the studied drug both during and a year after the NIPMS, with an increase of 6%–8%, though the impact was slightly smaller after the NIPMS had ended. In addition, they were more likely to prescribe the studied drug rather than alternative drugs used for the same indication during the NIPMS. This led to a tendency (albeit non-significant) toward higher revenue being generated for pharmaceutical companies by participating physicians’ prescriptions. None of the marketing indicators (i.e., indicators of NIPMSs being used for marketing purposes) proposed by our group in an earlier study were useful to predict whether an NIPMS would have a larger or smaller impact on prescription practices [9].

The reasons for the difference in prescription habits between participant physicians and controls cannot be assessed in this study. The difference may be due to a higher awareness of the studied drug because of participation in the NIPMS. Whether the remuneration offered for the inclusion of patients in the study plays a role is unclear, but the amount of remuneration and whether it is appropriate with respect to the amount of effort for the physician does not seem to be associated with the difference. The prescription of the studied drugs also increased compared to alternative drugs used for the same indication. This suggests that the difference is not due to increased diagnosis of the disease that the drug is used to treat, but rather due to a shift in prescription behavior towards the studied drug. Thus, patients with similar disorders are likely to be treated differently by a physician participating in an NIPMS compared to one not participating in an NIPMS. Although this study did not attempt to assess the appropriateness of medical prescriptions, the data nonetheless raise questions about the independence of physicians when prescribing drugs. Physicians participating in NIPMSs showed a higher prescription rate of the drug under study even before the start of the study, even though this was one of the matching criteria. We believe this may be due to the fact that physicians participating in an NIPMS may have already been in contact with representatives of the sponsor of the NIPMS and thus may have already been more aware of the drug compared to controls.

Strengths and weaknesses

One strength of this study is its large sample size, studying close to 7,000 physicians within a diverse collection of NIPMSs. It is thus highly likely that the results are generalizable to other NIPMSs and other physicians. The study design was quasi-experimental, allowing for assessment of causality when certain assumptions are met. However, controlling confounders was difficult due to data privacy laws in Germany. It is unclear whether physician age, gender, or specialization may have an influence on prescription habits. However, as mentioned in the Methods section, we used directed acyclic graph modeling to assess the effects of confounders, and believe it is unlikely we insufficiently controlled for these characteristics. Only in rare cases where a studied drug gains an indication during the time of the NIPMS could specialization lead to additional confounding, when one specialty would prescribe the studied drug for the new indication while another specialty would not. One other confounder that may not be sufficiently controlled for is that physicians may have chosen to participate in an NIPMS because they were already aware of a certain medication and actively wanted to gather experience using it, leading to an overestimation of the impact. In our view, however, this is not very likely because pharmaceutical companies are more likely to recruit physicians for NIPMSs who are not as enthusiastic about their medication yet [27].

Another weakness is the imprecise definition of the time frame of physician participation in the NIPMSs. Due to a lack of information regarding when exactly a physician entered or exited an NIPMS, we assumed in the primary analysis that all physicians had participated for the entire time of the NIPMS. This may lead to an under- or overestimation of the difference where we miscalculated the time frame of participation. However, our sensitivity analyses with the subset of physicians for whom a more precise time frame could be determined confirmed the difference we found in the larger set of physicians. It is thus unlikely that the difference would be changed substantially if precise data were available for all physicians.

Relation to other studies

To our knowledge, so far no other study has studied the impact of NIPMSs on participants’ prescription practices. Previous studies have focused primarily on the scientific quality of NIPMSs or the quality of the registrations [4,5,8]. The comparison with trials investigating interventional studies’ effects on prescription practices is difficult. For 2 trials identified as seeding trials, quantitative data are not available in sufficient quality to compare with the results of our current study [13,14]. Andersen et al. conducted an independent study of the effects of an interventional trial on physicians’ prescription practices and found increases in prescription habits comparable with those found in our current study, though slightly larger [15]. Due to the interventional nature of the trial, it is not surprising that it may have had a more pronounced impact on prescription practices than the NIPMSs in our study. Glass studied relative grant amounts from pharmaceutical companies to physicians participating in phase III trials and found no correlation between the relative grant amount and the subsequent prescription behavior of participating physicians, in line with our result that the appropriateness of remuneration for the NIPMS was not associated with the difference in prescription behavior [28].

Meaning and implications

Our study shows that participating in a “non-interventional” study may still lead to a change in prescription behavior of the participating physicians. This adds to the large body of evidence indicating that conflicts of interest resulting from interactions between physicians and the pharmaceutical industry influence physician behavior [2933]. It is unclear whether the change in prescription behavior resulting from NIPMSs is in the best interest of the patient, but currently neither the physicians nor the patients participating in such studies are being informed about it at all. More importantly, NIPMSs are currently subject to less scrutiny than interventional trials due to the assumption that they do not impact physicians’ prescriptions and thus do not result in patients being treated differently; for example, it is not required to acquire informed consent from a patient before enrolling them in an NIPMS. Our study casts strong doubts on this assumption. In addition, we were not able to show that certain marketing characteristics of an NIPMS are able to predict whether it impacts physician prescribing behavior. We have to thus assume that it will not be possible to regulate NIPMSs in a way that reduces their impact on prescribing behavior. This leads to our conclusion that NIPMSs should only be permitted when they are imposed by regulatory authorities or registered with a scientifically sound study design that allows for the collection of essential data.

Supporting information

S1 Appendix. Selection of alternative drugs.

(DOCX)

S2 Appendix. Matching methods.

(DOCX)

S3 Appendix. Validation process.

(DOCX)

S4 Appendix. Marketing.

(DOCX)

S1 Fig. Directed acyclic graph model of possible confounders.

(TIF)

S2 Fig. Flowchart of exclusion of participating physicians and controls.

(TIF)

S1 File. Request to German Federal Ministry of Health.

Request regarding the use of privacy-protected data for the purpose of this study.

(DOCX)

S2 File. STROBE statement.

(DOC)

S1 Table. Reasons for exclusions of NIPMSs.

(DOCX)

S2 Table. Detailed information on characteristics of NIPMSs.

(DOCX)

S3 Table. Sensitivity analysis 1.

Relative prescription rates of participating doctors versus controls considering only studied drugs manufactured by the sponsor (model for t0 adjusted for overall prescription rate; models for t1 and t2 adjusted for overall prescription rate and prescription rate of studied drug at t0).

(DOCX)

S4 Table. Sensitivity analysis 2.

Relative prescription rates of participating doctors versus controls using only data for physicians with a precise time frame of NIPMS participation (model for t0 adjusted for overall prescription rate; models for t1 and t2 adjusted for overall prescription rate and prescription rate of studied drug at t0).

(DOCX)

S5 Table. Sensitivity analysis 3.

Relative prescription rates of participating doctors versus controls using only data for physicians with a precise time frame of NIPMS participation and considering only drugs manufactured by the sponsor (model for t0 adjusted for overall prescription rate; models for t1 and t2 adjusted for overall prescription rate and prescription rate of studied drug at t0).

(DOCX)

S6 Table. Alternative drugs for each NIMPS.

Note that only alternative drugs that shared the first 3 places of the ATC code with the studied drug were included. For some studied drugs, alternative medications with a different ATC code may exist. In addition, for some alternative drugs, the dosage form was restricted to ensure comparability.

(DOCX)

Acknowledgments

We thank Andrea Appel, Ellen Gugelfuß, and Fide Marten for support in the data collection phase of the study. We thank Christian Fischer at the Innungskrankenkasse for help in the organization of the data validation process.

Abbreviations

DDD

defined daily dose

NIPMS

non-interventional post-marketing study

Data Availability

Data cannot be shared publicy due to data privacy laws in Germany and the requirement by the German Ministry of Health to erase the data because of concerns of data privacy. The raw data is held by the GKV-Spitzenverband, the National Association of Statutory Health Insurance Funds, https://www.gkv-spitzenverband.de/, Reinhardtstrasse 28, 10117 Berlin, Germany. Access to the data is subject to data privacy restrictions.

Funding Statement

The authors received no specific funding for this work.

References

  • 1.BMJ. Post-marketing observational studies: my experience in the drug industry. BMJ. 2012;344:e3990 10.1136/bmj.e3990 [DOI] [PubMed] [Google Scholar]
  • 2.Gesetz über den Verkehr mit Arzneimitteln (Arzneimittelgesetz—AMG): § 4 Sonstige Begriffsbestimmungen. 1976 Aug 24. Berlin: Federal Ministry of Justice and Consumer Protection; 1976. [cited 2020 Jun 3]. Available from: https://www.gesetze-im-internet.de/amg_1976/__4.html. [Google Scholar]
  • 3.European Medicines Agency. Post-authorisation safety studies (PASS). Amsterdam: European Medicines Agency; 2018. [cited 2019 Sep 30]. Available from: https://www.ema.europa.eu/en/human-regulatory/post-authorisation/pharmacovigilance/post-authorisation-safety-studies-pass-0. [Google Scholar]
  • 4.von Jeinsen BKJG, Sudhop T. A 1-year cross-sectional analysis of non-interventional post-marketing study protocols submitted to the German Federal Institute for Drugs and Medical Devices (BfArM). Eur J Clin Pharmacol. 2013;69(7):1453–66. 10.1007/s00228-013-1482-z [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Dietrich ES. Die meisten deutschen Anwendungsbeobachtungen sind zur Generierung wissenschaftlich valider Erkenntnisse nicht geeignet. Pharmacoeconomics Ger Res Articles. 2009;7(1):3–14. [Google Scholar]
  • 6.Sox HC, Rennie D. Seeding trials: just say “no.” Ann Intern Med. 2008;149(4):279 10.7326/0003-4819-149-4-200808190-00012 [DOI] [PubMed] [Google Scholar]
  • 7.London AJ, Kimmelman J, Carlisle B. Research ethics. Rethinking research ethics: the case of postmarketing trials. Science. 2012;336(6081):544–5. 10.1126/science.1216086 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Spelsberg A, Prugger C, Doshi P, Ostrowski K, Witte T, Hüsgen D, et al. Contribution of industry funded post-marketing studies to drug safety: survey of notifications submitted to regulatory agencies. BMJ. 2017;356:j337 10.1136/bmj.j337 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Koch C, Appel AS, Lieb K, Lubner SM, Kölbel R. Sind Anwendungsbeobachtungen ein Marketing-Tool?. Medizinrecht. 2018;36(4):225–31. [Google Scholar]
  • 10.Hasford J, Lamprecht T. Company observational post-marketing studies: drug risk assessment and drug research in special populations—a study-based analysis. Eur J Clin Pharmacol. 1998;53(5):369–71. 10.1007/s002280050395 [DOI] [PubMed] [Google Scholar]
  • 11.Wieseler B, McGauran N, Kaiser T. New drugs: where did we go wrong and what can we do better? BMJ. 2019;366:l4340 10.1136/bmj.l4340 [DOI] [PubMed] [Google Scholar]
  • 12.Woloshin S, Schwartz LM, White B, Moore TJ. The fate of FDA postapproval studies. N Engl J Med. 2017;377(12):1114–7. 10.1056/NEJMp1705800 [DOI] [PubMed] [Google Scholar]
  • 13.Krumholz SD, Egilman DS, Ross JS. STEPS: a narrative account of a gabapentin seeding trial. Arch Intern Med. 2011;171(12):1100–7. 10.1001/archinternmed.2011.241 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Hill KP, Ross JS, Egilman DS, Krumholz HM. The ADVANTAGE Seeding Trial: a review of internal documents. Ann Intern Med. 2008;149(4):251–8. 10.7326/0003-4819-149-4-200808190-00006 [DOI] [PubMed] [Google Scholar]
  • 15.Andersen M, Kragstrup J, Sondergaard J. How conducting a clinical trial affects physicians’ guideline adherence and drug preferences. JAMA. 2006;295(23):2759–64. 10.1001/jama.295.23.2759 [DOI] [PubMed] [Google Scholar]
  • 16.Gesetz über den Verkehr mit Arzneimitteln (Arzneimittelgesetz—AMG): § 67 Allgemeine Anzeigepflicht. 1976 Aug 24. Berlin: Federal Ministry of Justice and Consumer Protection; 1976. [cited 2019 Jul 23]. Available from: https://www.gesetze-im-internet.de/amg_1976/__67.html. [Google Scholar]
  • 17.Hennessy S, Bilker WB, Berlin JA, Strom BL. Factors influencing the optimal control-to-case ratio in matched case-control studies. Am J Epidemiol. 1999;149(2):195–7. 10.1093/oxfordjournals.aje.a009786 [DOI] [PubMed] [Google Scholar]
  • 18.Wacholder S, Silverman DT, McLaughlin JK, Mandel JS. Selection of controls in case-control studies. III. Design options. Am J Epidemiol. 1992;135(9):1042–50. 10.1093/oxfordjournals.aje.a116398 [DOI] [PubMed] [Google Scholar]
  • 19.Verband der Ersatzkassen. Daten zum Gesundheitswesen: Versicherte. Berlin: Verband der Ersatzkassen; 2020. [cited 2020 Apr 8]. Available from: https://www.vdek.com/presse/daten/b_versicherte.html. [Google Scholar]
  • 20.Wissenschaftliches Institut der AOK. ATC-Klassifikation für den deutschen Arzneimittelmarkt. Berlin: Wissenschaftliches Institut der AOK; 2019. [cited 2019 Oct 1]. Available from: https://www.wido.de/publikationen-produkte/arzneimittel-klassifikation/. [Google Scholar]
  • 21.Greenland S, Pearl J, Robins JM. Causal diagrams for epidemiologic research. Epidemiology. 1999;10(1):37–48. [PubMed] [Google Scholar]
  • 22.Richardson DB, Langholz B. Background stratified Poisson regression analysis of cohort data. Radiat Environ Biophys. 2012;51(1):15–22. 10.1007/s00411-011-0394-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Turner H, Firth D. Generalized nonlinear models in R: an overview of the gnm package. Comprehensive R Archive Network; 2018. [Google Scholar]
  • 24.Bürkner P-C. brms: an R package for Bayesian multilevel models using Stan. J Stat Softw. 2017;80(1):1–28. 10.18637/jss.v080.i01 [DOI] [Google Scholar]
  • 25.Højsgaard S, Halekoh U, Yan J. The R package geepack for generalized estimating equations. J Stat Softw. 2005;15(1):1–11. 10.18637/jss.v015.i02 [DOI] [Google Scholar]
  • 26.R Development Core Team. R: a language and environment for statistical computing. Version 3.6.2. Vienna: R Foundation for Statistical Computing; 2008. Available from: http://www.R-project.org. [Google Scholar]
  • 27.Trilling T. Pharmamarketing: ein Leitfaden für die tägliche Praxis. 3rd edition Berlin: Springer; 2015. [Google Scholar]
  • 28.Glass HE. Do clinical grant payment practices in phase 3 clinical trials influence subsequent clinical investigator prescribing behavior? Dis Manag. 2004;7(1):77–87. 10.1089/109350704322919014 [DOI] [PubMed] [Google Scholar]
  • 29.Austad KE, Avorn J, Franklin JM, Campbell EG, Kesselheim AS. Association of marketing interactions with medical trainees’ knowledge about evidence-based prescribing: results from a national survey. JAMA Intern Med. 2014;174(8):1283–90. 10.1001/jamainternmed.2014.2202 [DOI] [PubMed] [Google Scholar]
  • 30.Fleischman W, Agrawal S, King M, Venkatesh AK, Krumholz HM, McKee D, et al. Association between payments from manufacturers of pharmaceuticals to physicians and regional prescribing: cross sectional ecological study. BMJ. 2016;354:i4189 10.1136/bmj.i4189 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Yeh JS, Franklin JM, Avorn J, Landon J, Kesselheim AS. association of industry payments to physicians with the prescribing of brand-name statins in Massachusetts. JAMA Intern Med. 2016;176(6):763–8. 10.1001/jamainternmed.2016.1709 [DOI] [PubMed] [Google Scholar]
  • 32.Lieb K, Scheurich A. Contact between doctors and the pharmaceutical industry, their perceptions, and the effects on prescribing habits. PLoS ONE. 2014;9(10):e110130 10.1371/journal.pone.0110130 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33.Lundh A, Lexchin J, Mintzes B, Schroll JB, Bero L. Industry sponsorship and research outcome. Cochrane Database Syst Rev. 2017;2:MR000033 10.1002/14651858.MR000033.pub3 [DOI] [PMC free article] [PubMed] [Google Scholar]

Decision Letter 0

Clare Stone

24 Mar 2020

Dear Dr. Koch,

Thank you very much for submitting your manuscript "Effect of physicians’ participation in non-interventional post-marketing studies on their prescription habits: A retrospective two-armed cohort study" (PMEDICINE-D-19-04307) for consideration at PLOS Medicine.

Your paper was evaluated by a senior editor and discussed among all the editors here. It was also discussed with an academic editor with relevant expertise, and sent to independent reviewers, including a statistical reviewer. The reviews are appended at the bottom of this email and any accompanying reviewer attachments can be seen via the link below:

[LINK]

In light of these reviews, I am afraid that we will not be able to accept the manuscript for publication in the journal in its current form, but we would like to consider a revised version that addresses the reviewers' and editors' comments. Obviously we cannot make any decision about publication until we have seen the revised manuscript and your response, and we plan to seek re-review by one or more of the reviewers.

In revising the manuscript for further consideration, your revisions should address the specific points made by each reviewer and the editors. Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments, the changes you have made in the manuscript, and include either an excerpt of the revised text or the location (eg: page and line number) where each change can be found. Please submit a clean version of the paper as the main article file; a version with changes marked should be uploaded as a marked up manuscript.

In addition, we request that you upload any figures associated with your paper as individual TIF or EPS files with 300dpi resolution at resubmission; please read our figure guidelines for more information on our requirements: http://journals.plos.org/plosmedicine/s/figures. While revising your submission, please upload your figure files to the PACE digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at PLOSMedicine@plos.org.

We expect to receive your revised manuscript by Apr 14 2020 11:59PM. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

We ask every co-author listed on the manuscript to fill in a contributing author statement, making sure to declare all competing interests. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. If new competing interests are declared later in the revision process, this may also hold up the submission. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT. You can see our competing interests policy here: http://journals.plos.org/plosmedicine/s/competing-interests.

Please use the following link to submit the revised manuscript:

https://www.editorialmanager.com/pmedicine/

Your article can be found in the "Submissions Needing Revision" folder.

To enhance the reproducibility of your results, we recommend that you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see http://journals.plos.org/plosmedicine/s/submission-guidelines#loc-methods.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

We look forward to receiving your revised manuscript.

Sincerely,

Clare Stone, PhD

Managing Editor

PLOS Medicine

plosmedicine.org

-----------------------------------------------------------

Requests from the editors:

Abstract – Please add dates of the study as well as geographic information. I would usually ask for summary demographic information to be added to the abstract and it isn’t really fully relevant here, but I still think some information like average age and sex ratio would be good. Also Please be more explicit about what the study’s limitations are as the final sentence of the ‘Methods and Findings’ section.

At this stage, we ask that you include a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract. Please see our author guidelines for more information: https://journals.plos.org/plosmedicine/s/revising-your-manuscript#loc-author-summary

Data – even though data can’t be shared by you, please provide the email or link where you requested the data from so others can do the same.

Refs in main text - Please use square brackets instead of rounded.

Formatting errors/ - Line 99, for example. There appear to be errors in the main text )Error! Bookmark not defined..

Did your study have a prospective protocol or analysis plan? Please state this (either way) early in the Methods section.

a) If a prospective analysis plan (from your funding proposal, IRB or other ethics committee submission, study protocol, or other planning document written before analyzing the data) was used in designing the study, please include the relevant prospectively written document with your revised manuscript as a Supporting Information file to be published alongside your study, and cite it in the Methods section. A legend for this file should be included at the end of your manuscript.

b) If no such document exists, please make sure that the Methods section transparently describes when analyses were planned, and when/why any data-driven changes to analyses took place.

c) In either case, changes in the analysis-- including those made in response to peer review comments-- should be identified as such in the Methods section of the paper, with rationale.

Please complete the CONSORT checklist and ensure that all components of CONSORT are present in the manuscript, including [how randomization was performed, allocation concealment, blinding of intervention, definition of lost to follow-up, power statement].

Comments from the reviewers:

Reviewer #1: The authors report the findings of a quasi-experimental study evaluating the effects of non-interventional post-marketing studies on prescribing volume rate of prescribing of the particular drug studied in the NIPM during and in the following year. They matched every intervention physician with 2 controls who were not participants in an NIPM study.

The study found that the relative rate of prescribing of physicians in NIPM studies were higher for particular drug studied during and in the year after.

The strengths of the study include its large study sample, attempt to control for confounding through matching by prescription volume of prescriptions and defined daily dosage in the year before, and the use of DAGs to conceptualise the potential causal pathways.

The use of a conditional Poisson regression analysis to determine the relative rates is appropriate type of statistical model for this type of analysis with an outcome related to count data.

Major comments

The primary weaknesses are due to its non-randomised design, only matching based on a very limited number of factors (due to availability of data due to privacy laws), confounding by indication and residual confounding (e.g. propensity for certain characteristics of physicians who may join a NIPM study in the first place are not taken into account), lack of a sample size calculation on a potential hypothesized effect size.

The authors could have also considered, for the experimental arm (physicians involved in NIPMS), a negative exposure drug which was the studies drug under question and conduct a sensitivity analyses to look at the relative rate of a non-studied drug compared to matched controls. Critically, if there still is an effect, it is likely that there will be residual confounders which were not considered in the primary analyses. If there is no effect - this would strengthen the conclusion of the paper. If there is an effect - unfortunately this would make their findings less than definitive.

Due to this - my feeling was the conclusion was too definitive given the limitations in the non-randomised design of the study as well as the short-time horizon (for instance, does this effect persist after the NIPM ends: the results in Table 3 show that at t1 the prescribing increases but then substantially falls towards t2 at 1 year). The authors' primary conclusion that NIPMS should be more strictly regulated due to changes in prescribing practices in favour of the investigated drug. However, the key issue that hasn't been investigated is whether any of these changes prescribing practices actually result in clinical benefit, clinical harm, or no change for the patient.

Minor comments:

Line 99 - check reference link, seems to be an error

Sample size calculation: There must be some rationale for 1:2 matching or else why not just run the analysis on 1:1 matching. In the discussion, the authors mention a previous trial (ref 15) Anderson et al. which did seem to have a measure of increase % of sales in an interventional trial. Could the authors not have used the trials figures to give an estimated samples size powered on an expected difference from the cited trial?

Table 3 results - Could the authors explain why there an increase for both control and NIPMS groups during t1 and the residual effects seems to reduce over time for both groups

Line 341 - Statement is factual incorrect - The study itself was not a trial design and quasi-experimental trials are not designed to specifically assess causality but rather this type design allows you to use casual inference methods when RCT are infeasible.

Lines 356 - Authors touch on this issue in the discussion, but how you deal confounding by indication that those physicians who also participate in NIPMs are more likely to sign up to medications may be interested in or have a preference for? One suggestion could be to consider a negative control drug (which is not studied in the NIPMS) for participating physician.

Reviewer #2: This is a fascinating article written on a highly important subject. The influence of industry in medical practice is of great concern globally. One particular way in which researchers have suspected that industry might try to influence physicians is through the conduct of post-marketing trials, which are ostensibly about generating new scientific information but are hypothesized to serve primarily marketing or "seeding" purposes. However, due to lack of uniform data across different companies, trials, and health care institutions, this hypothesis has been difficult to test. The authors of this study are to be commended for their application of a novel data source which appears to be exactly the kind of data needed to test this hypothesis.

I have several suggestions below to improve the manuscript. I will divide these into "major" and "minor" comments; in my view "major" comments are those that I would consider mandatory to be adequately addressed before publication. "Minor" comments are recommended, but may not be essential.

Major recommendations:

1. The manuscript is not clearly written. It would greatly benefit from the assistance of a medical writer to help edit for flow, style, and clarity. There are also several places in the manuscript, particularly the Methods, which need to be reorganized in order to convey more clearly what analyses were done.

2. Related to comment (1) above, I am still not 100% sure that I have fully understood the analytic approach that was taken in this paper. The analysis appears to be strong, but I would want to review the methodology again after it has been edited for clarity in order to make sure that this is the case.

3. Line 127 refers to "eligible NIMPS," which as of this point has not been defined yet. I would suggest starting the methods section with a full and clear description of inclusion of NIMPS before launching in to other details. This is in line with comment #1 above regarding need to reorganize the Methods section for greater clarity.

4. The analysis of prescribing during t1 needs to be described more fully. Periods t0 and t2 are both presented as being exactly 1 year in length. But t1 is not, and will be different for every NIMPS. Were prescriptions during t1 annualized? If so, did the authors account for uncertainty in estimates arising from shorter NIMPS being "inflated" up to a full year period?

5. Related to #4 above, the authors refer to prescribing "rates," but the results I see appear to be simply prescribing counts. A rate implies a "per unit time" in the measure. Are the results presented in Table 3 (for example) the number of prescriptions PER YEAR? If so, this would in fact be a rate…but this should be described more clearly.

6. More detail needs to be given regarding the "indicators of marketing." Because of the centrality of these measures to the current study, I don't think it is enough to refer only to the prior publication. It would be fine to put it in the supplement, but I think more description (are these measures objective? Subjective? What criteria were they based on?) needs to be present in the current manuscript.

7. The authors need to include results regarding the success of their matching algorithm. The central question that needs to be addressed here is how closely the exposed and control physicians were able to be matched on their prescription counts. For example, the difference in the number of prescriptions between the exposed and control physicians could be determined within each matched trio, and then the distribution of the difference across all matched trios could be shown in a supplementary table. Without presentation of data such as these (or some other analysis to the same effect), the apparent difference between exposed and control physicians in terms of drug prescribed during the t0 period is highly concerning.

8. I believe that physician specialty has a potential role as a confounder beyond that which is discussed by the authors. For some drugs, physician specialty could affect prescriptions at t1 independent of the effect at t0; this is because drugs are constantly gaining new indications and usages which may apply to only some specialties. For example, infliximab has longstanding uses for rheumatologic/inflammatory conditions, but increasing use in treating immunotherapy-related toxicity in cancer patients. Therefore, I would expect t0 = t1 for a rheumatologist or gastroenterologist, but t1 > t0 for an oncologist. The authors state that physician specialty is not available to them, so I think that specialty needs more discussion in Limitations as an unmeasured confounder.

9. My best understanding of the results is that lines 220-225 describe the analysis presented in table 4. However, lines 226-229 appear to describe a separate analysis, and it is not clear to me where these results are presented?

Minor recommendations:

1. Abstract, line 49 (and other locations): the comparison of different exposure groups is implicit within retrospective cohort studies. Therefore, the term "two arm" is not needed, and confusedly may be taken to imply that this was a randomized study.

2. Line 80: What does "pharmaceutical authorities" refer to? Drug companies? Regulatory bodies such as the FDA and the EMA? If the latter, "regulatory agencies" might be a better term.

3. Line 90: the phrase "the fact that NIPMS seem to be published extremely rarely" does not make sense. See major recommendation #1 above regarding need for style and clarity edits.

4. Line 99: "Error, bookmark not defined"

5. Lines 213-14: Authors state that trios without "at least one control" were excluded. Does this mean that those with only one control were still included? These would be pairs, not trios. Hence, "trio" might not be the best term if some of them were in fact 1:1 rather than 1:2.

6. Line 221: Explain further. How were coefficients for matching trios generated? Was this a fixed-effects model, with trio-level effects?

7. Table 3. How do the authors interpret the apparent increase in prescriptions among controls during t1? It does not seem to me that prescribing should increase at all in this group.

8. Line 288: Does this statistic (1.04 with 95% CI 1.03-1.05) refer back to the first row of table 4? If so, why is this statistic described as an Rpr in one place and an OR in the other? If not, what table does it refer to?

9. Throughout, it is not necessary to have sentences like "table X shows that…" Instead, it is clearer just to make the statement and then cite the referenced table at the end, as in (Table 3).

10. Line 335-337. This explanation does not seem adequate. If exposed physicians were already more interested in the study drug in t0, then they would have been matched to higher-prescribing controls in t0. Matching would seem to have accounted for this.

11. In discussion, consider comparing to findings by HE Glass, 2003: https://europepmc.org/article/med/15035835

12. The conclusion that NIPMS should be stopped is overly strong without a much fuller discussion of what the potential benefits and harms of these types of studies are.

Reviewer #3: [Major Points]

1. This authors' perspective is very unique. The article is worth publication with the condition that the following are described clearly.

(1) Page7, Line128: What is the "Overall prescription volume"? Does it mean that a number of prescriptions per the doctor, per the drug, or per a certain period (e.g. per a month)? And what the aim of using it for control matching.

(2) Page7, Line133-137: As for Definitions of the "indicators of marketing use", the author just refereed their previous report but it should be described in this paper because these are necessary to interpreter the result accurately. I was not able to access the report on pub-med as probably that is written in German.

(3) Page9, Line171: What is the purpose of setting two periods of each four months for gathering the physicians' identified data?

(4) Page10, Line189: What are included in the "revenue generated by these prescriptions"? Is it only remuneration for the INPMS or included sales profits etc.?

(5) Page8, Table1, 3rd Cullum of Inclusion criteria for NIPMS: Dose the "follow up of one year (t2)" mean one year immediately after the end of the INPMS period? If not, which period has acquired?

(6) Page8, Table1, 7th Cullum of Inclusion criteria for NIPMS: Author have set the "approved at least 6 months before the beginning of the NIPMS" but 12 months were required to be satisfied one year data before INPMS (t0).

(7) Page13, Table2: There are several NIPMS whose duration is over one year. What rules were applied to extract one-year prescription data from one-year or more data?

(8) S1 Appendix. Selection of alternative drugs: How has the author identified alternative drugs when there was more than one drug meets the ATC-code criteria? If specified alternative drugs for each NIPMS were stated in S2 Table Detailed information on characteristics of NIPMS, it would be helpful.

(9) S2 Appendix. Matching Methods: What the aim of taking into account the "similar defined daily dose (DDD)-prescription of the studied drug" for identifying controls? In my understanding it depends on drugs, not on physicians.

2. In the discussion, the bellow points should be considered.

(1) Table 3 shows that even the control has a higher number of packages and DDD on t1 than t0 and t2. What could explain this trend most reasonably?

(2) Table 4 shows that there is a statistical difference in t0, before INPMS. It seems to indicate that physicians participating in a NIPMS already had a preference for new drugs (studied drugs) over existing drugs (alternative drugs), regardless of experience on NIPMS.

(3) In S2 Table, there is a large variety of "Duration approval date until begin (mo)" on NIPMS. In general, a situation of using drugs is affected by the time after approval. How did it affect the results of this study?

(4) In S2 Table, No.1-6 NIPMS have less than 10 physicians. Is there any possibility that few physicians influence the results?

[Minor Point]

3. Page6, Line99: there is "Error! Bookmark not defined."

Reviewer #4: I commend the authors in examining the association of the effect of a potential marketing device disguised as a research study with drug prescription. The retrospective study was based on a fairly large number of physicians of nearly 7000 using what appears to a national claims database that found participation in the NIPMS was associated with higher rates of prescription of the studied medication in the periods during the participation in the NIPMS and the year after.

Some specific questions and issues to clarify:

1) In the introduction section- Participation in NIPMS. How do physicians actually "participate" in NIPMS? How are physicians typically "selected" to participate? This seem to imply the physicians participate in the design/analysis of the study as opposed to something more similar to be detailed (marketed) about the drug. What are the participants suppose to do and actually do?

2) In the methods section- National Association of Statutory Health Insurance Funds. Readers may not be familiar with this. Is this the name of the national claims database or prescription database? Or the name of agency? I assume this database includes all physician prescription database or only containing a subset of physicians or prescriptions? It seems the database only provide the prescription information for physicians in private practice. What is the implication of this? Are most physicians in private practice or work in the hospital? Is this referring to prescriptions in out-patient versus in-patient settings?

3) In the methods section- Study design on matching to controls. It is unclear to me (or at least that I can easily decipher) after reading through the methods section what criteria that authors used to match the participants in NIPMS and control. Matched on overall prescription volume at T0? What are included in the prescription volume? Could the patient's specialty impact the prescription volume? Matched to DDD and packages? Can the number of years out of training/medical school be found? Table 3- what is the prescription rate normalized to?

4) In the Methods section- NIPMS/ Table 2. The drugs studied is a very heterogenous class of drugs. Some have alternative equivalent. Some not as much (i.e. ivabradine). Some are much more prevalent/commonly used. In the analysis consider whether categories of class of drugs (cardiac, oncologic, ENT/primary care) affect the association of participation of NIPMS and prescription rate.

5) Analysis section of outcome measures- Have authors considered measuring "differences in difference" in the prescribing rates of the participants and controls in between the periods T0, T1, and T2?

Any attachments provided with reviews can be seen via the following link:

[LINK]

Decision Letter 1

Clare Stone

6 May 2020

Dear Dr. Koch,

Thank you very much for re-submitting your manuscript "Effect of physicians’ participation in non-interventional post-marketing studies on their prescription habits: A retrospective two-armed cohort study" (PMEDICINE-D-19-04307R1) for review by PLOS Medicine.

I have discussed the paper with my colleagues and the academic editor and it was also seen again by the original reviewers. I am pleased to say that provided the remaining editorial and production issues are dealt with we are planning to accept the paper for publication in the journal.

The remaining issues that need to be addressed are listed at the end of this email. Any accompanying reviewer attachments can be seen via the link below. Please take these into account before resubmitting your manuscript:

[LINK]

Our publications team (plosmedicine@plos.org) will be in touch shortly about the production requirements for your paper, and the link and deadline for resubmission. DO NOT RESUBMIT BEFORE YOU'VE RECEIVED THE PRODUCTION REQUIREMENTS.

***Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.***

In revising the manuscript for further consideration here, please ensure you address the specific points made by each reviewer and the editors. In your rebuttal letter you should indicate your response to the reviewers' and editors' comments and the changes you have made in the manuscript. Please submit a clean version of the paper as the main article file. A version with changes marked must also be uploaded as a marked up manuscript file.

Please also check the guidelines for revised papers at http://journals.plos.org/plosmedicine/s/revising-your-manuscript for any that apply to your paper. If you haven't already, we ask that you provide a short, non-technical Author Summary of your research to make findings accessible to a wide audience that includes both scientists and non-scientists. The Author Summary should immediately follow the Abstract in your revised manuscript. This text is subject to editorial change and should be distinct from the scientific abstract.

We expect to receive your revised manuscript within 1 week. Please email us (plosmedicine@plos.org) if you have any questions or concerns.

We ask every co-author listed on the manuscript to fill in a contributing author statement. If any of the co-authors have not filled in the statement, we will remind them to do so when the paper is revised. If all statements are not completed in a timely fashion this could hold up the re-review process. Should there be a problem getting one of your co-authors to fill in a statement we will be in contact. YOU MUST NOT ADD OR REMOVE AUTHORS UNLESS YOU HAVE ALERTED THE EDITOR HANDLING THE MANUSCRIPT TO THE CHANGE AND THEY SPECIFICALLY HAVE AGREED TO IT.

Please ensure that the paper adheres to the PLOS Data Availability Policy (see http://journals.plos.org/plosmedicine/s/data-availability), which requires that all data underlying the study's findings be provided in a repository or as Supporting Information. For data residing with a third party, authors are required to provide instructions with contact information for obtaining the data. PLOS journals do not allow statements supported by "data not shown" or "unpublished results." For such statements, authors must provide supporting data or cite public sources that include it.

If you have any questions in the meantime, please contact me or the journal staff on plosmedicine@plos.org.

We look forward to receiving the revised manuscript by May 13 2020 11:59PM.

Sincerely,

Clare Stone, PhD

Managing Editor

PLOS Medicine

plosmedicine.org

------------------------------------------------------------

Requests from Editors:

Title – please add a country setting and as below remove ‘effect’. I suggest a change from Effect of physicians’ participation in non-interventional post-marketing studies on their 2 prescription habits: A retrospective two-armed cohort study

To

Impact of physicians’ participation in non-interventional, post-marketing studies on their prescription habits: A retrospective two-armed cohort study in Germany

Data – thank you for clarifying that the data is not publicly available. Please however add the URL /email or address or contact for those – who like you did – would like to approach the ministry of Health for access to the data.

Abstract – please remove “Trial Registration: The trial was not registered, as it was not a clinical trial.”

As this is not a trial, please avoid saying "effect" (e.g. in the title; line 47; line 92; line 360 etc) based on this research design. "Impact" is the maximum that would be acceptable, I would think

Throughout manuscript - more careful about claims throughout, for example, you say "changes prescription rates" at lines 66 and 358 (however, all that is shown is that one group of physicians behave differently from a similar group)

Please briefly tell us about the classes of drugs studied in the abstract, e.g., X of 24 studies involved oncology agents, Y cardiovascular etc.

Is it possible to say something about the geography of this study – which cities? If too many to mention, maybe just a brief outline of some information – avoiding a list. Just to give a sense of relevance.

You mention "patient harm" at line 97; I don't think there is any data on patient harm, is there? If so, I don't think you should imply that harm might result, rather saying, at most, "we have no data on patient benefits or harms".

STROBE checklist – please use sections and paragraphs instead of pages as these change during revisions and formatting, etc.

Comments from Reviewers:

Reviewer #1: Thanks for the authors detailed response to my review. Their additional explanations are very helpful in helping clarifying some of my understanding, in particular how the authors minimised the effect of confounding for voluntary NIPMS participation by physicians.

In the response to this they have provided good justification on assessing matching criteria on the relative difference between cases and control during time period t0 as well as controlling for differences in the Poisson regression models. I think their approach is pragmatic and it would be strengthen their arguments, if they could detail this in their manuscript

For the SS calculation - this is fine as a justification that they erred on a conservative approach to try to sample larger numbers as they felt a trial results may lead to under-powering their study.

One final comment is that the results clearly show that there is increase in prescribing behaviour for the therapeutic products being involved in NIPMS, therefore it would most likely make sense that there might decrease rates in therapeutic products for the same indication which patients could be switched off from. Would the authors comment on whether they believe this effect could be observed?

Reviewer #2: Major comments:

1. Lines 186-188. Were the matching criteria (1) number of overall prescriptions in packages and (2) DDD of the studied drug and (3) DDD of alternative drugs? Or, (1) number of overall prescriptions in packages and (2) DDD of (the studied drug + alternative drugs)? Or something else? This should be clearer. Additionally, were both of these criteria (or all three, if that is the case) factored in equally to the matching?

2. Related to #1 above, but also to my prior comment on the initial submission (Reviewer 2, minor recommendation #10), I am still not clear how, in a strictly mathematical/statistical sense, the t0 differences in prescribing presented in table 4 can be present if groups have been matching on t0 prescribing. If matching occurred based on (overall prescriptions) and (studied drug + alternative drugs), then it seems plausible that such a difference could arise if NIPMS physicians had at baseline a greater % of studied drug within the studied drug + alternative drug total. However, if matching used (overall prescriptions) and (studied drug) and (alternative drugs) separately, then it is not clear to me how the differences in table 4 persisted. Especially since it does not appear that the models were adjusted for other potential confounders, besides these prescribing measures?

Minor comments:

1. Line 120. I believe the authors are making a valid point about publication bias here. However, the phrasing of "NIMPS are also published extremely rarely" might be confused by the reader to mean "NIMPS are also conducted extremely rarely." I would suggest rephrasing to make it clear the NIMPS are often conducted but rarely published.

2. Line 177-178. As phrased, it sounds as though the authors had to get approval from the individual NIPMS in order to get data on physician prescribing practices, which I do not expect was the intended meaning.

Reviewer #3: The authors dedicated to clarifying the method of the study by revising it but there are still unclear points especially in the results. To help readers understand the manuscript without misinterpretation, I recommend additional revisions.

[Major Points]

1. Page 16, Line 307, "7-8% more of the studied drug during the NIPMS and 6-7% more during the year after the NIPMS" The figures have one digit after the decimal point, and how to calculated these percentages from the number of prescriptions in Table 3.

2. Page 18, Line 329, I'm not sure why the author examines only during t1 (NIPMS) not t2 (immediate after NIPMs) time frame for proportions of prescription of the studied drugs compared to alternative drugs. But this fact should be explicitly stated on page 20, line 362 of the discussion section.

3. Page 18, Table 5, The percentage in parentheses do not match the figures calculated by dividing each number of NIPMS by 24. Besides I don't understand what the author intended to explain from "not appropriate: 4 (14.8%); unclear whether appropriate: 6 (22.2%)."

4. Page 18, Table 5, The statements of marketing indicators should be described in the same terms as on the S4 Appendix and S2 Table.

[Minor Points]

5. Page 17, Table 3, I understood the periods are different among the t0, t1. If the average period of each time frame is shown in Table 3, it would be easily recognized the difference.

6. Before publication the author should look though details of the manuscript since there are several typos (i.e. lack of space; page 2, line 58, "NIPMSprescribed," page 16, line 304, "prescriptionsof").

Any attachments provided with reviews can be seen via the following link:

[LINK]

Decision Letter 2

Clare Stone

27 May 2020

Dear Dr. Koch,

On behalf of my colleagues and the academic editor, Dr. Sanjay Basu, I am delighted to inform you that your manuscript entitled "Impact of physicians’ participation in non-interventional post-marketing studies on their prescription habits: A retrospective two-armed cohort study in Germany" (PMEDICINE-D-19-04307R2) has been accepted for publication in PLOS Medicine.

PRODUCTION PROCESS

Before publication you will see the copyedited word document (in around 1-2 weeks from now) and a PDF galley proof shortly after that. The copyeditor will be in touch shortly before sending you the copyedited Word document. We will make some revisions at the copyediting stage to conform to our general style, and for clarification. When you receive this version you should check and revise it very carefully, including figures, tables, references, and supporting information, because corrections at the next stage (proofs) will be strictly limited to (1) errors in author names or affiliations, (2) errors of scientific fact that would cause misunderstandings to readers, and (3) printer's (introduced) errors.

If you are likely to be away when either this document or the proof is sent, please ensure we have contact information of a second person, as we will need you to respond quickly at each point.

PRESS

A selection of our articles each week are press released by the journal. You will be contacted nearer the time if we are press releasing your article in order to approve the content and check the contact information for journalists is correct. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact.

PROFILE INFORMATION

Now that your manuscript has been accepted, please log into EM and update your profile. Go to https://www.editorialmanager.com/pmedicine, log in, and click on the "Update My Information" link at the top of the page. Please update your user information to ensure an efficient production and billing process.

Thank you again for submitting the manuscript to PLOS Medicine. We look forward to publishing it.

Best wishes,

Clare Stone, PhD

Managing Editor

PLOS Medicine

plosmedicine.org

Associated Data

    This section collects any data citations, data availability statements, or supplementary materials included in this article.

    Supplementary Materials

    S1 Appendix. Selection of alternative drugs.

    (DOCX)

    S2 Appendix. Matching methods.

    (DOCX)

    S3 Appendix. Validation process.

    (DOCX)

    S4 Appendix. Marketing.

    (DOCX)

    S1 Fig. Directed acyclic graph model of possible confounders.

    (TIF)

    S2 Fig. Flowchart of exclusion of participating physicians and controls.

    (TIF)

    S1 File. Request to German Federal Ministry of Health.

    Request regarding the use of privacy-protected data for the purpose of this study.

    (DOCX)

    S2 File. STROBE statement.

    (DOC)

    S1 Table. Reasons for exclusions of NIPMSs.

    (DOCX)

    S2 Table. Detailed information on characteristics of NIPMSs.

    (DOCX)

    S3 Table. Sensitivity analysis 1.

    Relative prescription rates of participating doctors versus controls considering only studied drugs manufactured by the sponsor (model for t0 adjusted for overall prescription rate; models for t1 and t2 adjusted for overall prescription rate and prescription rate of studied drug at t0).

    (DOCX)

    S4 Table. Sensitivity analysis 2.

    Relative prescription rates of participating doctors versus controls using only data for physicians with a precise time frame of NIPMS participation (model for t0 adjusted for overall prescription rate; models for t1 and t2 adjusted for overall prescription rate and prescription rate of studied drug at t0).

    (DOCX)

    S5 Table. Sensitivity analysis 3.

    Relative prescription rates of participating doctors versus controls using only data for physicians with a precise time frame of NIPMS participation and considering only drugs manufactured by the sponsor (model for t0 adjusted for overall prescription rate; models for t1 and t2 adjusted for overall prescription rate and prescription rate of studied drug at t0).

    (DOCX)

    S6 Table. Alternative drugs for each NIMPS.

    Note that only alternative drugs that shared the first 3 places of the ATC code with the studied drug were included. For some studied drugs, alternative medications with a different ATC code may exist. In addition, for some alternative drugs, the dosage form was restricted to ensure comparability.

    (DOCX)

    Attachment

    Submitted filename: Response_to_reviewers.docx

    Attachment

    Submitted filename: Response_to_reviewers_2.docx

    Data Availability Statement

    Data cannot be shared publicy due to data privacy laws in Germany and the requirement by the German Ministry of Health to erase the data because of concerns of data privacy. The raw data is held by the GKV-Spitzenverband, the National Association of Statutory Health Insurance Funds, https://www.gkv-spitzenverband.de/, Reinhardtstrasse 28, 10117 Berlin, Germany. Access to the data is subject to data privacy restrictions.


    Articles from PLoS Medicine are provided here courtesy of PLOS

    RESOURCES