We read with great interest the longitudinal study by Xie et al on the Risk of death among users of Proton Pump Inhibitors (PPI) (1). The study findings suggest that there is an overall increased risk of death among PPI users when compared with histamine H2 receptor antagonists (H2 blockers) users and people who are not on either medication. We also acknowledge the efforts of the authors to statistically eliminate potential bias associated with an observational study design like this. However, we would like to highlight a few points which we believe should be considered when interpreting the results of the study.

The authors externally adjusted for the use of other therapeutics including anticoagulants, antiplatelet agents, and non-steroidal anti-inflammatory drugs as potential confounders. However, it is not clear if Selective Serotonin Receptor Inhibitors (SSRIs) used in the treatment of Post-Traumatic Stress Disorder (PTSD) were considered and adjusted for. The Veteran population which forms the data set for this study has a higher prevalence of PTSD than the general population and is often prescribed SSRIs (2). The SSRIs are metabolized by the cytochrome P-450 isoenzyme CYP2C19, which is inhibited by the proton pump inhibitors (PPIs) leading to significantly higher levels of SSRIs in the body and potential risk of increased adverse effects (3). Some of the adverse effects of SSRIs including suicide risk could significantly contribute to mortality in the population studied, especially in the presence of co-morbidities like cardiovascular diseases.

Though the study demonstrated a linear graded association between duration of PPI exposure and risk of death among new users, we believe the gradient effect cannot be effectively ascertained without analyzing the dose of PPIs that was actually taken by the study subjects. The study cohort consists mostly of the elderly, and evidence has shown increased risk of accidental prescription drug abuse and overdose in this population (4 - 6). PPIs, just like every other medication, if used above the intended dose could cause deleterious adverse effects. Therefore, without the data on the dose of PPI among users, it might be inaccurate to associate duration of use in isolation with increased risk of mortality.

As expected of an observational study, it is difficult to validate cause and effect. The cause of death in the study population is not known and the mortality noted could possibly be secondary to underlying medical conditions and not necessarily treatment with PPIs. Figure 1 from the study analysis showed that an estimated 10% of the population on PPIs died within 1 year (1). As the authors mentioned, the biologic mechanism behind the observed mortality risk from PPI use is not clear. Nevertheless, one would expect the latency period between PPI exposure and any mortality risk to be more than 1 year. Again, this strongly suggests the deaths may not entirely be linked to treatment with PPI as there could be a significant contribution from underlying medical problems.

The study analysis suggests an increased mortality risk with PPI use compared with H2 blocker use. However, the severity of medical conditions in the H2 blocker and PPI groups was not accounted for. Taking cardiovascular disease, for instance, the authors controlled for the presence of heart disease in both groups and not severity. It is quite possible that people with mild heart disease could have been prescribed an H2-blocker while people with severe heart disease were put on a PPI. This would create a bias towards an increased risk of mortality in the PPI group which is not related to use as suggested by the researchers.

Overall, we appreciate the authors for a well-conducted research. In light of evidence from other studies linking PPI use to a host of medical problems, findings from this study further emphasize the need for judicious use of PPI. Healthcare providers should regularly review prescriptions with patients and constantly provide education on the risks of continued use of medications without any clinical indication. Lastly, more studies should be conducted on the risk of PPIs using a database that is more representative of the general population to enable generalizability of study findings.

References
1. Xie Y, Bowe B, Li T, Xian H, Yan Y, Al-aly Z. Risk of death among users of Proton Pump Inhibitors : a longitudinal observational cohort study of United States veterans. 2017;1–12.
2. Alexander W. Pharmacotherapy for Post-traumatic Stress Disorder In Combat Veterans: Focus on Antidepressants and Atypical Antipsychotic Agents. P T. 2012;37(1):32–8.
3. Gjestad C, Westin AA, Skogvoll E, Spigset O. Effect of proton pump inhibitors on the serum concentrations of the selective serotonin reuptake inhibitors citalopram, escitalopram, and sertraline. Ther Drug Monit. 2015;37(1):90–7.
4. Hernandez SH, Nelson LS. Prescription Drug Abuse: Insight Into the Epidemic. Clin Pharmacol Ther. 2010;88(3):307–17.
5. Seniors and Prescription Drug Addiction by AgingCare( 04/17/12, http://www.agingcare.com/Articles/Seniors-and-Prescription-Drug-Addictio...)
6. Elderly at Risk for Prescription Drug Abuse by The Partnership at DrugFree.org( 04/17/12, http://www.drugfree.org/join-together/addiction/elderly-at-risk-for-pres...)

Dear Editor,

We thank Drs Burt, Rasminsky, Andrade, and Barros for their letters on our study published in BMJ Open in January 2017.1 We agree with them that our baseline prevalence of major malformations in the Quebec Pregnancy Cohort (QPC) is higher than the reported 3-5%. Bérard et Sheehy2 have described the QPC, and have presented baseline statistics showing that the baseline prevalence of major malformations in Quebec was 9.3%. We have disclosed that already in our Discussion section – ‘Our major congenital malformation population prevalence may seem somewhat higher than the routinely reported 3-5%, but in fact our rate is consistent with what is expected in the province of Quebec, due to high concentration of genetic risk factors stemming from the ‘founding’ French ancestors.’1,3 It is true that we are referencing another of our studies to highlight the Founders’ effect. Indeed, Zhao et al.3 have shown that the prevalence of major malformations in the province of Quebec vary by geographical regions, and are higher in the regions where the founding French ancestors first inhabited. This is a well know fact and is representative of specific genetic traits of those founding fathers. Again, this was well discussed and represented in Zhao et al.,3 and it is why we are referencing this study in Bérard et al.1 Bérard et al.1 further showed that the prevalence of major malformations among those with depression during pregnancy was even higher (greater than 11%). However, we do not believe that this higher prevalence of major malformations biased our findings. Indeed, we have already mentioned this in our Discussion section – ‘Nevertheless, given that the baseline rate of major congenital malformations is similarly higher in those taking antidepressants vs non-users in our study, it has no impact on the internal validity of our study because it cancels out when comparing the exposed and unexposed groups.’

Finally, we have studied depressed pregnant women, some of which were treated with antidepressants. This was done to take into account indication bias per design as was explained, hence increasing internal validity. We have further discussed external validity in our Discussion section – ‘as we considered pregnant women insured by the prescription drug insurance program, generalizability of results to those insured by private drug insurance could be affected. However, validation studies have shown that pregnant women receiving medication insurance from Quebec’s public system have similar characteristics and co-morbidities to those who have private medication insurance. Although socio-economic status (SES) might differ between the two groups, it does not affect internal validity given that all have similar SES status. Similarly, although the study population slightly differed from pregnancies not meeting eligibility criteria in terms of SES, this does not affect interval validity.’1,4

1. Bérard A, Zhao JP, Sheehy O. Antidepressant use during pregnancy and the risk of major congenital malformations in a cohort of depressed pregnant women: an updated analysis of the Quebec Pregnancy Cohort. BMJ Open. 2017 Jan 12;7(1):e013372.
2. Bérard A, Sheehy O. The Quebec Pregnancy Cohort--prevalence of medication use during gestation and pregnancy outcomes. PLoS One. 2014 Apr 4;9(4):e93870.
3. Zhao JP, Sheehy O, Bérard A. Regional variations in the prevalence of major congenital malformations in Quebec: The importance of fetal growth environment. J Popul Ther Clin Pharmacol. 2015;22(3):e198-210.
4. Bérard A, Lacasse A. Validity of perinatal pharmacoepidemiologic studies using data from the RAMQ administrative database. Can J Clin Pharmacol. 2009 Summer;16(2):e360-9.

We acknowledge of course that the pharmaceutical industry produces many medicines that are an important part of healthcare. However, inappropriate use and overuse of medicines is detrimental to health, and use of an unnecessarily expensive medicine that is no better than a cheaper alternative represents wasteful health care spending.

Although Mr. Catelin states “It’s ludicrous to suggest that a sandwich and a soda water would sway the opinions of medical practitioners”, the evidence shows otherwise. Among the studies that have examined the relationship between exposure to pharmaceutical industry information or payments and the quality, quantity and cost of prescribing, there is an overwhelming finding of no benefit. (1-4) Pharmaceutical industry payments, even small payments such as a meal, are associated with increased prescriptions of the promoted drug. One US study found a “dose-response effect” on prescribing of cholesterol-lowering drugs, blood pressure drugs and antidepressants, starting with even a single meal costing less than US $20: the more free meals a doctor ate, the more likely they were to prescribe the promoted drug. (2) Another study found that the more payments a doctor received from the industry, the more likely they were to prescribe expensive brand-name drugs rather than less expensive generic equivalents. (4)

The aim of our study was neither to undermine patients’ confidence nor to call health professionals into disrepute. The aim was to provide public access to information on the extent of industry funding of professionals in Australia, and to stimulate an informed discussion on how to ensure that patients and clinicians base decisions on the best available evidence from independent sources.

Finally, following the lead of other countries such as the US and France, we recommend that Medicines Australia report all payments to health professionals, including food and beverages of nominal value.

We appreciate the opportunity to continue the discussion on these topics.

Alice Fabbri, Quinn Grundy, Barbara Mintzes, Swestika Swandari, Ray Moynihan, Emily Walkom, Lisa Bero

References:

1. Yeh JS, Franklin JM, Avorn J, et al. Association of Industry Payments to Physicians With the Prescribing of Brand-name Statins in Massachusetts. JAMA Intern Med 2016;176(6):763-8.
2. DeJong C, Aguilar T, Tseng C, et al. Pharmaceutical Industry-Sponsored Meals and Physician Prescribing Patterns for Medicare Beneficiaries. JAMA Intern Med 2016;176(8):1114-10.
3. Spurling G, Mansfield P, Montgomery BD, et al. Information from pharmaceutical companies and the quality, quantity, and cost of physicians' prescribing: a systematic review. PLoS Med 2010;7(10):e1000352.
4. Ornstein C, Grochowski R, Tigas M. Now There’s Proof: Docs Who Get Company Cash Tend to Prescribe More Brand-Name Meds. Pro Publica: March 17, 2016. Available at: https://www.propublica.org/article/doctors-who-take-company-cash-tend-to.... Accessed July 7, 2017.

Dear Tom,

Thank you for your comments.

Your first point addresses the poor negative predictive value of a TB symptom screening test. We agree that the symptom screen that is currently called for in South Africa’s national guidelines, as well as WHO’s guidelines, is inadequate in terms of both negative and positive predictive value. We do not have, and you do not propose, however, an easily-available alternative. We consulted a number of TB experts in designing the algorithm and concluded that at the time of study initiation, there was no readily available clinical prediction rule or point-of-care instrument that we could expect to be adopted at a national scale if the trial’s results are compelling. We also designed the study to adhere as closely as possible to existing national guidelines, with the main difference being the timing of the initiation process rather than its content. Our goal is an algorithm that can be incorporated into routine practice as simply as possible, in the hope of benefiting patients at a large scale.

We also observe that initiating ART in the presence of undiagnosed TB poses the risk of delayed TB treatment (which would have been delayed anyhow, due to the patient not having symptoms) or IRIS (which, while clinically occasionally challenging, is rarely fatal, especially in this group of ambulatory and largely healthy patients). We hypothesize that the well-characterized loss to follow-up and subsequent morbidity and mortality due to delays in ART initiation will outweigh any benefit from an intensive investigation for sub-clinical TB (which you do not define, but is likely to be complex and not inexpensive, if utilising technology with high sensitivity such as Xpert MTB/RIF). We think the algorithm can easily be adapted to accommodate future objective tests that may supplement or replace the screening tests. However, even in the absence of these tests, we believe that a basic symptom screen that can be performed by nurses and requires no additional technology has high utility.

Regarding the sample size, you are correct that a high proportion of those who screen out of immediate ART initiation using SLATE will receive standard of care treatment and therefore are not likely to have improved outcomes compared to our control group. We anticipate the numbers who screen out will be lower than you suggest, and we calculated our original sample size using the best available data at that time. Still, you are correct that if the proportion screening out is higher than anticipated, we will have less power than anticipated. We recently increased our sample size in South Africa to 660 to allow for detection of smaller differences should they occur and to improve our ability to analyze those who screen out as a population in themselves.

Next, you suggest that we could shift randomization to after the SLATE screen. This would defeat the purpose of the algorithm entirely, as the point of the study is to test whether we can effectively screen out those who should not start on ART immediately, screen in provide efficient initiation to those who can, and improve overall outcomes for the population of patients who are eligible for ART (i.e. nearly all HIV-positive individuals not already on ART). If we took your suggestion and randomized after screening, we would be asking a very narrow question that lacks policy relevance—namely, does immediate ART initiation among a selected population that is already at low risk for poor outcomes have better outcomes if offered same day treatment compared to standard of care. We don’t believe this is the right question to ask.

Finally, with regard to your conclusion, we are responding to a pressing need for guidance on same day initiation, and we used WHO and other symptom screen algorithms and national guidelines to construct that algorithm. Algorithms are often constructed from imperfect science, and as you note, the perfect algorithm is not available. We would argue we have thoughtfully weighed the potential for harm against the need for public health clinics to move on a pressing health problem, and the only harm we can see in our approach (other than perhaps screening too many patients out with the screening tools, which does not disadvantage these patients, merely returns them to standard care) is a slightly increased risk of IRIS, which may in fact not differ from that in standard care.

We do agree that a better algorithm is needed. We are already discussing opportunities to improve on SLATE. The results of the SLATE study, when available, will guide us on the right way to go.