Co-morbidity	Avoid	Effect
Parkinson’s disease	Neuroleptics	Exacerbates Parkinsonian symptoms (including tremor)
Hypertension	Non-steroidal anti-inflammatory drugs	Sodium retention
Asthma	Beta-blockers, adenosine	Bronchospasm
Respiratory failure	Morphine, diamorphine	Respiratory depression
Renovascular disease	ACE inhibitors/antagonists	Reduction in glomerular filtration
Chronic heart failure	Trastuzumab	Worsening of heart failure
Chronic infections (e.g. tuberculosis, hepatitis C, histoplasmosis	Cytokine modulators (e.g. etanercept)	Increased risk of exacerbation

Prescribing in neonates, infants, children and adolescents

The use of drugs in young people poses special problems. Extrapolating from adult dosage regimens, merely adjusting for weight, leads to excessive (and potentially toxic) doses because:

The rates of hepatic metabolism and renal excretion of drugs are reduced in neonates and infants.

Premature babies have approximately 1% of their body weight as fat (compared with 20% in adults), leading to a marked increase in plasma drug levels of fat-soluble drugs.

There are other difficulties in prescribing for children:

Many treatments have never been subject to formal trials in either children or adolescents and their benefits and risks have not, therefore, been appropriately assessed in these age groups. Efforts are being made, internationally, to redress this.

For many drugs, there are no paediatric preparations or formulations. Instead, adult products are used.

Precise oral dosing is often impossible in babies who spit out unpleasant-tasting products!

Adverse effect profiles of medicines may be different in children compared with adults (e.g. Reye’s syndrome in children given aspirin, suicidal ideas in depressed adolescents treated with selective serotonin reuptake inhibitors).

Prescribing for the elderly

The use of drugs in the elderly is often a problem:

Rates of hepatic drug metabolism and renal excretion decline with age. Extrapolation of drug dosages, from those appropriate in younger adults, may therefore lead to toxic plasma levels.

Changes in drug distribution due to changes in body composition, and the preferential distribution of the cardiac output to the brain, may also predispose to toxicity.

Co-morbidity, often associated with polypharmacy, leads to increased opportunities for disease–drug and drug–drug interactions.

Concordance with treatment regimens diminishes as the number of prescribed drugs increases, and is especially poor in the face of cognitive impairment.

Exaggerated pharmacodynamic effects of drugs acting on the central nervous, cardiovascular and gastrointestinal systems are common.

Examples of common problems encountered in the use of drugs amongst older people are shown in Table 17.2.

Drug use in pregnancy

Clinicians should be extremely cautious about prescribing drugs to pregnant women, and only essential treatments should be given. When a known teratogen is needed during pregnancy (e.g. an anticonvulsant drug or lithium), the potential adverse effects should be discussed with the parents, preferably before conception. If parents wish to go ahead with the pregnancy, they should be offered an appropriate ultrasound scan to assess whether there is any fetal damage. Some known human teratogens are shown in Table 17.3.

Table 17.3 Some human teratogens

Drug	Effect
ACE inhibitors/antagonists	Oligohydramnios
Retinoids, e.g. acitretin	Multiple abnormalities
Carbimazole	Neonatal hypothyroidism
	Abnormalities of bone growth
Antiepileptics
Carbamazepine	Cleft palate
Lamotrigine
Phenytoin
Valproate	Neural tube defects
NSAIDs	Delayed closure of the ductus arteriosus
Cytotoxic drugs	Most are presumed teratogens
Lithium	Ebstein’s anomaly
Misoprostol	Moebius’s syndrome
Thalidomide (and possibly lenalidomide)	Phocomelia

Note: All drugs should be avoided in pregnancy unless benefit clearly outweighs the risk.

Breast-feeding

Although most drugs can be detected in breast milk, the quantity is generally small. This is because, for most drugs, the concentration in milk is in equilibrium with plasma water (i.e. the non-protein-bound fraction). A few drugs (e.g. aspirin, carbimazole) may, however, cause harm to the infant if ingested in breast milk. Relevant drug literature should be consulted when prescribing for nursing mothers.

The drug

Selecting the right drug involves three elements:

The drug’s clinical efficacy for the proposed use

The balance between the drug’s efficacy and safety

Patient preference.

The most common approach to assessing a drug’s efficacy is the randomized controlled trial (RCT), although other approaches (see p. 907) can be informative. The demonstration of absolute efficacy (against placebo) may, itself, be insufficient. Where there is more than one treatment for the same indication these should be compared with one another, taking account of the magnitude of their benefits, their individual adverse reaction profiles, and their costs.

Direct comparisons of one treatment versus another are particularly useful but are often unavailable. So-called indirect techniques, which involve comparing each drug against placebo and then imputing the comparison, are commonly used.

Patients’ own preferences should be discussed to enable them to be equal partners in decision-making about whether, and how, they wish to be treated. Moreover, a full understanding of the reasons for considering treatment, the likely benefits and the possible adverse reactions, has repeatedly been shown to improve ‘concordance’ with treatment regimens.

The dose

Appropriate drug dosages will have usually been determined from the results of so-called ‘dose-ranging’ studies during the original development programme. Such studies are generally conducted as RCTs covering a range of potential doses. Drug doses and dosage regimens may be fixed or adjusted.

Fixed dosage regimens

Drugs suitable (in adults) for prescribing at the same ‘fixed’ dose, for all patients, share common features. Efficacy is optimal in virtually all patients; and the risks of dose-related (type A) adverse reactions (see p. 904) are normally low. These drugs have a high ‘therapeutic ratio’ (i.e. the ratio between toxic and therapeutic doses). Examples of drugs prescribed at a fixed dose are shown in Table 17.4.

Table 17.4 Examples of fixed dose prescribing

Drug	Indication
Aspirin	Secondary prevention of myocardial infarction
Clopidogrel	Secondary prevention of myocardial infarction
Bendroflumethiazide	Hypertension
Broad spectrum penicillins	Lower urinary tract infection
Cephalosporins	Lower urinary tract infection
Macrolides	Upper and lower respiratory tract infection
Levonorgestrel	Emergency contraception
Ulipristal	Emergency contraception
Oestrogen antagonists	Secondary prevention of breast cancer
Aromatase inhibitors	Secondary prevention of breast cancer
Vaccines	e.g. Diphtheria, pertussis, mumps, measles, rubella, influenza, etc.

Titrated dosage regimens

For many drugs, there are wide interindividual variations in response. As a consequence, whilst a particular dose may in one person lack any therapeutic effect, the same dose in another may cause serious toxicity. The reasons for such variability are partly due to pharmacokinetic factors (differences in the rates of drug absorption, distribution or metabolism) and partly due to pharmacodynamic factors (differences in the sensitivity of target organs).

Pharmacokinetics

Pharmacokinetics is the study of what the body does to a drug. The intensity of a drug’s action, immediately after parenteral administration, is largely a function of its volume of distribution. This, in turn, is predominantly governed by body composition and regional blood flow. Dosage adjustments, for body weight or surface area, are therefore common (e.g. in cancer chemotherapy) in order to optimize treatment.

The main determinants of a drug’s plasma concentration after oral administration are its bioavailability (the proportion of the unchanged drug that reaches the systemic circulation) and its rate of systemic clearance (by hepatic metabolism or renal excretion). A drug’s oral bioavailability depends on the extent to which it is:

destroyed in the gastrointestinal tract

able to cross the gastrointestinal epithelium

metabolized by the liver before reaching the systemic circulation (so-called presystemic or ‘first pass’ metabolism). First pass metabolism can be avoided by the intravascular (i.v.), intramuscular (i.m.) or sublingual routes.

Liver drug metabolism occurs in two stages:

Phase I is the modification of a drug, by oxidation, reduction or hydrolysis. Of these, oxidation is the most frequent route and is largely undertaken by a family of isoenzymes known as the cytochrome P450 system (see p. 902). Inhibition or induction of cytochrome P450 isoenzymes are major causes of drug interactions (Table 17.5).

Phase II involves conjugation with glucuronate, sulphate, acetate or other substances to render the drug more water soluble and therefore able to be excreted in the urine.

Table 17.5 Some inducers and inhibitors of cytochrome P450

Inducers	Carbamazepine
	Hyperforin^a
	Nifedipine
	Non-nucleoside reverse transcriptase inhibitors (NNRTIs)
	Omeprazole
	Phenobarbital
	Phenytoin
	Rifampicin
	Ritonavir (see p. 180)
Inhibitors	Allopurinol
	Amiodarone
	Cimetidine
	Erythromycin, clarithromycin
	Fluoxetine, paroxetine
	Grapefruit juice (contains flavonoids)
	Imidazoles
	Quinolones
	Saquinavir
	Sulphonamides

a Hyperforin is one of the ingredients of the herbal product known as St John’s wort used by herbalists to treat depression. Although it is marketed as a licensed medicine, it is a reminder that drug interactions can occur with alternative, as well as conventional, medicines.

Genetic causes of altered pharmacokinetics

Both presystemic hepatic metabolism, and the rate of systemic hepatic clearance, may vary markedly between healthy individuals.

Variability in the genes that encode drug-metabolizing enzymes (Table 17.6) is a major determinant of the inter-individual differences in the therapeutic and adverse responses to drug treatment. The most common involve polymorphisms of the cytochrome P450 family of enzymes, CYP. The first to be discovered was the polymorphism in the hydroxylation of the antihypertensive agent debrisoquin (CYP2D6). Defective catabolism was shown to be a monogenetically inherited trait, involving 5–10% of Caucasian populations, and leading to an exaggerated hypotensive response.

Table 17.6 Some genetic polymorphisms involving drug metabolism

Enzyme	Drug
P450
Cytochrome CYP1A2	Amitriptyline
Cytochrome CYP1A2	Clozapine
Cytochrome CYP3A4	Amlodipine
	Ciclosporin
	Nifedipine
	Sildenafil
	Simvastatin
	Protease inhibitors
	Tacrolimus
Cytochrome CYP2C9	Warfarin
	Glipizide
	Losartan
	Phenytoin
Cytochrome CYP2D6	Beta-blockers
	Codeine
	Risperidone
	SSRIs
	Tramadol
	Venlafaxine
Cytochrome CYP2C19	Clopidogrel^a
	Cyclophosphamide
	Diazepam
	Lansoprazole
	Omeprazole
Plasma pseudocholinesterase	Succinylcholine
Plasma pseudocholinesterase	Mivacurium
Thiopurine methyltransferase	Azathioprine
Thiopurine methyltransferase	Mercaptopurine
UDP-glucuronosyl transferase	Irinotecan
N-acetyl transferase	Isoniazid

CYP, cytochrome; SSRIs, Selective serotonin reuptake inhibitors.

a Clopidogrel is a prodrug and impaired metabolizers have a reduced response.

A substantial number of other drugs – estimated at 15–25% of all medicines in use – are substrates for CYP2D6. The frequencies of the variant alleles show racial variation and a small proportion of individuals may have two or more copies of the active gene. The phenotypic consequences of the defective CYP2D6 include the increased risk of toxicity with those antidepressants or antipsychotics that undergo metabolism by this pathway. Conversely, in individuals with multiple copies of the active gene, there are extremely rapid rates of metabolism and therapeutic failure at conventional doses.

Warfarin is predominantly metabolized by CYP2C9. In most populations, between 2% and 10% are homozygous for an allele that results in low enzyme activity. Such individuals will therefore metabolize warfarin more slowly leading to higher plasma levels, a greater risk of bleeding, and a requirement for lower doses if the international normalized ratio (INR) is to be maintained within the therapeutic range.

Cytochrome P450 is inhibited by the proton pump inhibitors but the consensus view is that co-prescribing with clopidogrel does not cause a significant increase in cardiovascular risk.

Individual differences in the activity of thiopurine methyltransferase (TPMT) determine the doses of mercaptopurine and azathioprine that are used. TMPT activity is therefore undertaken routinely in children undergoing treatment for acute lymphatic leukaemia and people with Crohn’s disease (see p. 233).

Many drugs undergo metabolism by more than one member of the cytochrome P450 family. Individuals deficient in one enzyme may have normal, or over-expressed, activities of others. Current knowledge (and cost) does not therefore permit predictions of an individual’s dosage requirements for the wide range of drugs for which polymorphisms in metabolism have been identified.

This may, however, become possible in the future, and would contribute – in part – to the prospect of ‘personalized prescribing’ (see p. 899).

Other causes of altered pharmacokinetics

Rates of hepatic drug clearance can also be influenced by environmental factors including diet, alcohol consumption and concomitant therapy with drugs capable of inducing or inhibiting (Table 17.5) drug metabolism. Hepatic drug clearance also decreases with age. By contrast, renal drug clearance does not show substantial variation between healthy individuals although it declines with age and in people with intrinsic renal disease.

Pharmacodynamics

Pharmacodynamics is the study of what the drug does to the body. Pharmacodynamic sources of variability in the intensity of drug action are at least partly due to drug receptor polymorphisms (Table 17.7). At present, the pharmacodynamic tests used in clinical practice to target therapy are largely confined to the expression of:

oestrogen and HER2 receptors in women with breast cancer (to determine, respectively, responsiveness to anti-oestrogens and trastuzumab)

epidermal growth factor receptor in lung cancer and glioblastomas (to predict responsiveness to gefitinib).

The prospect for ‘personalized prescribing’ will be enhanced further, when pharmacodynamic polymorphisms can be elicited by gene scanning. The interplay between pharmacokinetics and pharmacodynamics will then permit drug selection and dosing to become much more precise.

Monitoring the effects of treatment

The combination of pharmacokinetic and pharmacodynamic causes of variability makes monitoring of the effects of treatment essential. Three approaches are used.

Affordability

The money available for healthcare varies widely across the world and there are marked differences (Fig. 17.1). All healthcare systems try to provide their populations with the highest standards of care within the resources they have at their disposal. The expenditure of large sums on a few people may deprive many of cost-effective remedies – a phenomenon known as the ‘opportunity cost’. The differences in healthcare expenditure shown in Figure 17.1 can be very largely accounted for by their differences in national wealth as reflected by their gross domestic products.

Figure 17.1 Annual expenditure on healthcare, as US$ per head of the population, in some developed countries.

(Source: OECD; http://www.oecd.org/document/4/0,3746,en_2649_37407_35101892_1_1_1_37407,00.html.)

In many countries cost-containment measures are encouraged (or mandated). For example, to reduce costs, all drugs should be prescribed by their generic (approved) names rather than their ‘brand’ ones because, once their patents have expired, generics products are cheaper. Despite occasional claims to the contrary, generic products are required to go through the same stringent regulatory processes as their branded counterparts.

Some countries, including Australia, Canada and Britain, assess the cost-effectiveness of new drugs (value for money) before they are available under their publicly funded healthcare systems.

Adverse drug reactions

Adverse drug reactions (ADRs), defined as ‘the unwanted effects of drugs occurring under normal conditions of use’, are a significant cause of morbidity and mortality. Around 5% of acute medical emergencies are admitted with ADRs, and around 10–20% of hospital inpatients suffer an ADR during their stay. Unwanted effects of drugs are five to six times more likely in the elderly than in young adults; and the risk of an ADR rises sharply with the number of drugs administered.

Classification

Two types of ADR are recognized.

Type A (augmented) reactions (Table 17.9) are:

qualitatively normal, but quantitatively abnormal, manifestations of a drug’s pharmacological or toxicological properties

predictable from its known pharmacological or toxicological actions

generally dose-dependent

usually common

only occasionally serious.

Table 17.9 Examples of adverse drug reactions

Type of reaction and drug	Adverse reaction
Type A (augmented)
ACE inhibitors	Hypotension Chronic cough
ACE antagonists	Hypotension
Anticoagulants	Gastrointestinal bleeding Intracerebral haemorrhage
Antipsychotics	Acute dystonia/dyskinesia Parkinsonian symptoms Tardive dyskinesia
Cytotoxic agents	Bone marrow dyscrasias Cancer
Erythromycin	Nausea, vomiting
Glucocorticosteroids	Hypoadrenalism Osteoporosis
Insulin	Hypoglycaemia
Tricyclic antidepressants	Dry mouth
Type B (bizarre)
Benzylpenicillin Radiological contrast media	Anaphylaxis
Broad-spectrum penicillins	Maculopapular rash
Carbamazepine Lamotrigine Phenytoin Sulphonamides	Toxic epidermal necrolysis Stevens–Johnson syndrome
Carbamazepine Diclofenac Isoniazid Phenytoin Rifampicin	Hepatotoxicity
Isotretinoin SSRIs^a	Depression Suicidal ideation

ACE, angiotensin-converting enzyme; SSRIs, selective serotonin reuptake inhibitors.

a In children and adolescents.

Whilst some such reactions as hypotension with ACE inhibitors may occur after a single dose, others may develop only after months (pulmonary fibrosis with amiodarone) or years (second malignancies with anti-cancer drugs).

Type B (idiosyncratic) reactions (Table 17.9) have no resemblance to the recognized pharmacological or toxicological effects of the drug. They are:

qualitatively abnormal responses to the drug

unpredictable from its known pharmacological or toxicological actions

generally dose-independent

usually rare

often serious.

Diagnosis

All ADRs mimic some naturally occurring disease and the distinction between an iatrogenic aetiology, and an event unrelated to the drug, is often difficult. Although some effects are obviously iatrogenic (e.g. acute anaphylaxis occurring a few minutes after intravenous penicillin), many are less so. There are six characteristics that can help distinguish an adverse reaction from an event due to some other cause:

Appropriate time interval. The time interval between the administration of a drug and the suspected adverse reaction should be appropriate. Acute anaphylaxis usually occurs within a few minutes of administration, whilst aplastic anaemia will only become apparent after a few weeks (because of the life-span of erythrocytes). Drug-induced malignancy, however, will take years to develop.

Nature of the reaction. Some conditions (maculopapular rashes, angio-oedema, fixed drug eruptions, toxic epidermal necrolysis) are so typically iatrogenic that an adverse drug reaction is very likely.

Plausibility. Where an event is a manifestation of the known pharmacological property of the drug, its recognition as a type A adverse drug reaction can be made (e.g. hypotension with an antihypertensive agent, or hypoglycaemia with an antidiabetic drug). Unless there have been previous reports in the literature, the recognition of type B reactions may be very difficult. The first cases of depression with isotretinoin, for example, were difficult to recognize as an ADR even though a causal association is now acknowledged.

Exclusion of other causes. In some instances, particularly suspected hepatotoxicity, an iatrogenic diagnosis can only be made after the exclusion of other causes of disease.

Results of laboratory tests. In a few instances, the diagnosis of an adverse reaction can be inferred from the plasma concentration (Table 17.8). Occasionally, an ADR produces diagnostic histopathological features. Examples include putative reactions involving the skin and liver.

Results of dechallenge and rechallenge. Failure of remission when the drug is withdrawn (i.e. ‘dechallenge’) is unlikely to be an ADR. The diagnostic reliability of dechallenge, however, is not absolute: if the ADR has caused irreversible organ damage (e.g. malignancy) then dechallenge will result in a false-negative response. Rechallenge, involving re-institution of the suspected drug to see if the event recurs, is often regarded as an absolute diagnostic test. This is, in many instances, correct but there are two caveats. First, it is rarely justifiable to subject a patient to further hazard. Second, some adverse drug reactions develop because of particular circumstances which may not necessarily be replicated on rechallenge (e.g. hypoglycaemia with an antidiabetic agent).

Management

As a general rule, type A reactions can usually be managed by a reduction in dosage whilst type B reactions almost invariably require the drug to be withdrawn (and never re-instituted).

Specific therapy is sometimes required for ADRs such as bleeding with warfarin (vitamin K), acute dystonias (benztropine) or acute anaphylaxis (see Emergency Box 3.1, p. 69).

Evidence-based medicine

There is general acceptance that clinical practice should, as far as possible, be based on evidence of benefit rather than theoretical speculation, anecdote or pronouncement.

One of the main applications of ‘evidence-based medicine’ is in therapeutics. Treatments should be introduced into, and used in, routine clinical care only if they have been demonstrated to be effective in appropriate studies. Three approaches are used:

1. Randomized controlled trials

2. Controlled observational trials

3. Uncontrolled observational studies.

Randomized controlled trials

In this type of study, people with a particular condition are allocated to one of two (and sometimes more) treatments randomly. At the end of the study, the outcomes in the groups are compared. The purpose of the randomized controlled trial is to minimize bias and confounding. In order to minimize patient bias, the patients themselves are generally unaware of their treatment allocations (a ‘single-blind’ trial); and in order to reduce doctor bias, treatment allocations are also withheld from the investigators (a ‘double-blind’ trial). To recruit sufficient numbers of patients, and to examine the effects of treatment in different settings, it is often necessary to conduct the trial at several locations (a ‘multicentre’ trial).

Randomized controlled trials are designed to assess whether one treatment is better than another (a ‘superiority’ trial); or whether one treatment is similar to another (an ‘equivalence’ trial).

In a superiority trial the study treatment is usually compared to placebo or to current standard practice.

In an equivalence trial the treatment under study is compared to another treatment for the same condition.

Although RCTs were originally introduced to investigate the efficacy of drugs, the methodology can be used for surgical (and other) procedures and medical devices.

There are a number of variants of the conventional randomized controlled trial including cross-over trials, cluster randomized controlled trials, inferiority trials and futility trials (see Further Reading).

Assessing randomized controlled trials

In assessing the relevance and reliability of an RCT a number of features need to be taken into account.

Randomization

In any RCT the method of randomization should be robust. In particular, the investigator should be unaware of which treatment a patient entering a trial will receive. This avoids selection bias.

Maintaining blindness

Although, ideally, in RCTs neither the investigator nor the patient is aware of the treatment allocation until the end of the study, this is not always possible. Adverse drug reactions, for example, may make it obvious which treatment a patient has been given. Nevertheless maintaining ‘blindness’ is necessary where the outcome is subjective (e.g. relief of pain, alleviation of depression) if bias is to be avoided.

Were the treated and control groups comparable?

Were they similar in their ‘baseline’ characteristics? Were they, for example, of similar age, severity and duration of illness? If not, are the differences likely to influence the results? Has the statistical analysis (using analysis of covariance, or Cox’s proportional hazards model) (see below) tried to adjust for them? Table 17.10 shows some of the baseline characteristics of a trial comparing prednisolone with placebo in the treatment of Bell’s palsy (idiopathic facial paralysis).

Table 17.10 Summary of a multicentre randomized placebo-controlled trial of prednisolone (25 mg twice daily), for 10 days, in the treatment of Bell’s palsy

Outcomes

There are two ways to look at the outcomes of an RCT.

Per protocol analysis: this includes only those who completed the study.

Intention-to-treat analysis: this includes all patients from the time of randomization.

Ideally, there should be no difference but in reality the results of a per protocol analysis are usually more advantageous to a treatment than an intention-to-treat analysis. The reason is that the intention-to-treat analysis will take account of patients who have withdrawn from the trial because of intolerance of the treatment or adverse drug reactions. It is therefore a much more robust approach. The results of the intention-to-treat analysis, in the trial of prednisolone in Bell’s palsy, are shown in Table 17.10. The trial results indicate, with a high probability, that treatment of Bell’s palsy with prednisolone will increase the chances of a full recovery of facial nerve function.

Are the results generalizable?

Were the patients enrolled into the study a reasonable reflection of those likely to be treated in routine clinical practice (a so-called pragmatic trial)? Or were they a selected population that excluded significant patient groups (such as the elderly)? If the latter, view the results with caution.

Analysis of a superiority trial

The analysis of a superiority trial is based on the premise – the ‘null hypothesis’ – that there is no difference between the treatments. The null hypothesis is rejected if the probability of the observed result occurring by chance, the p value, is less than 1 in 20 (i.e. p < 0.05). There are three caveats.

Any difference may still be due to chance; and it is often better to await the results of at least two independent studies before adopting a new treatment.

A trial may show no ‘statistically significant’ difference, when one in fact exists, because too few patients have been included, in other words the trial lacked sufficient ‘power’. The ‘power’ of a study (the number of patients needed in each treatment group to detect a predefined difference) should have been defined at the outset. If the study was underpowered, the results of the study should be interpreted with extreme care.

A statistically significant difference may not, necessarily, be clinically relevant.

Scrutiny of the magnitude of the effect, and its 95% confidence intervals (CI), is a far better guide than the p value.

Effect size. The results of the well-designed trial in Table 17.10 show, very convincingly, that the treatment of Bell’s palsy with prednisolone increases the chances of complete recovery of facial nerve function, at 12 months, from 81.6% to 94.4%. This is a far more convincing description of the benefits of treatment than the p value.

Another expression of the benefit of a treatment such as prednisolone can be derived from the number needed to treat (NNT). This is an estimate of the numbers of patients needed to be treated with a drug to achieve one positive result. In the study shown in Table 17.10, the NNT to enable one patient with Bell’s palsy to regain normal facial nerve function, after prednisolone treatment, is eight.

Analysis of an equivalence trial

The aim of an equivalence trial is to determine whether two (or possibly more) treatments produce similar benefits. During the design of such trials, it is necessary to decide what difference is unimportant and then to calculate the number of patients needed in order to have an 80% or 90% chance of showing this. In equivalence trials such power calculations show that the number of patients required is invariably greater than those needed for superiority trials. In such studies the comparator itself must, of course, already have been shown to be effective.

Meta-analysis

It is possible to summate all the controlled trials that have been performed in the treatment of a particular condition so as to refine the estimate of effectiveness. This technique minimizes random error in the assessment of the effect size of a treatment because more patients are included than could be accommodated in any single trial. A meta-analysis should be performed (and interpreted) carefully because of the heterogeneity of the individual studies used in it.

Controlled observational trials

Three types of observational study have been used to test the clinical effectiveness of therapeutic interventions:

Historical controlled trials

Case–control studies

Before-and-after studies.

Historical controlled trials

Despite the value of the prospective randomized controlled trial there are many treatments that have never been subjected to this technique, yet their efficacy is unquestioned. Examples include insulin in the treatment of diabetic ketoacidosis, thyroxine for hypothyroidism, vitamin B₁₂ in pernicious anaemia and defibrillation for ventricular fibrillation. In a historical controlled trial the outcome in patients treated with the study drug is compared to that of previously untreated people with the same disease. Treatments can be accepted into routine use on the basis of favourable comparisons with historical controls when the following criteria are met:

There should be a biologically plausible basis for the observed benefits.

There should be no appropriate treatment that could be reasonably used as a control.

The condition should have a known and predictable natural history.

The treatment should not be expected to have adverse effects that would compromise its potential benefits.

There should be a reasonable expectation that the magnitude of the therapeutic effects of treatment will be large enough to make the interpretation of its benefits unambiguous.

Case–control studies

This type of study design compares people with a particular condition (the ‘cases’) with those without (the ‘controls’). The approach has predominantly been used to identify epidemiological ‘risk factors’ for specific conditions such as lung cancer (smoking) or sudden infant death syndrome (lying prone); or in the evaluation of potential adverse drug reactions (such as deep venous thrombosis with oral contraceptives).

A case–control design allows an estimation of the odds ratio (OR), which is the ratio of the probability of an event occurring to the probability of the event not occurring (Box 17.1).

Box 17.1

Estimation of odds ratio

	Cases	Controls
Risk factor present	a	c
Risk factor absent	b	d
The odds ratio (OR) = (a ÷ b) / (c ÷ d)

An OR that is significantly greater than unity indicates a statistical association that may be causal. The OR for deep venous thrombosis and current use of oral contraceptives equals 2–4 (depending on the preparation): this indicates that the risk of developing a deep venous thrombosis on oral contraceptives is between 2 and 4 times greater than the background rate.

In some studies, the OR for a particular observation has been found to be significantly less than unity, suggesting ‘protection’ from the condition under study. Some studies of women with myocardial infarction indicated protection in those using hormone-replacement therapies but it has been subsequently shown that the result was due to bias. On the other hand, case–control studies have consistently shown that aspirin and other non-steroidal anti-inflammatory drugs are associated with a reduced risk of colon cancer. This seems to be a causal effect.

Case–control studies claiming to demonstrate the efficacy of a drug need to be interpreted with great care: the possibility of bias and confounding is substantial as was seen in the studies of hormone-replacement therapy and myocardial infarction. Confirmation from one or more RCTs is usually essential.

Before-and-after studies

It has sometimes been inferred that observed improvements seen in patients before, and after, the application of a particular treatment is evidence of efficacy. Such an approach is fraught with difficulties: the combination of a placebo effect, as well as regression to the mean, is likely to negate most studies using this type of design. Nevertheless, there are some circumstances where genuine efficacy can be confidently observed with such designs: the consequences of hip replacement, and cataract surgery, are good examples. Such instances can be regarded as special examples of the use of implicit historical controls.

Uncontrolled observational studies

Uncontrolled case series cannot be considered as providing primary evidence of efficacy unless they are undertaken in circumstances that are virtually those of historical controlled trials. When used in this way to demonstrate clinical effectiveness their validity relies on the use of implicit historical trials. Case series can, however, sometimes be of value in demonstrating the generalizability of the results of RCTs.

Dangers

Drug trials are carried out in specific groups of selected patients under strict supervision. The results, particularly when dramatic, are often used outside the strict inclusion criteria for clinical trials. The dramatic effect of spironolactone in heart failure (30% reduction in all-cause mortality) has not always been replicated in routine clinical practice because the wrong patients have been treated, often with higher doses, leading to hyperkalaemia and death.

Evaluation of new drugs

New drugs are subjected to a vigorous programme of preclinical and clinical testing before they are licensed for general use (Table 17.11) and are also monitored for safety following licensing. Doctors are recommended to fill in yellow cards when they suspect an adverse reaction has taken place.

Table 17.11 Evaluation of new drugs

Phase I: Healthy human subjects (usually men)

First use in man

Evaluation of safety and toxicity

Pharmacokinetic assessment

Sometimes pharmacodynamic assessment

Approximately 100 subjects

Phase II: First assessment in patients

Safety and toxicity evaluated

Dose range identified

Pharmacokinetic and pharmacodynamic monitoring

Approximately 500 subjects

Phase III: Use in wider patient population

Efficacy main objective

Safety and toxicity also carefully monitored

Often multicentre trials

Approximately 2000 patients involved

Phase IV: Post-marketing surveillance

All patients prescribed drug monitored

Efficacy, safety and toxicity measured

Quantification of unusual drug adverse effects

Yellow Card and Prescription Event Monitoring

Often very large numbers of patients observed

Statistical analyses

The relevance of statistics is not confined to those who undertake research but also to anyone who wants to understand the relevance of research studies to their clinical practice.

The average

Clinical studies may describe, quantitatively, the value of a particular variable (e.g. height, weight, blood pressure, haemoglobin) in a sample of a defined population. The ‘average’ value (or ‘central tendency’ in statistical language) can be expressed as the mean, median or mode depending on the circumstances:

Averages. In symmetrical data the mean, mode and median are the same.

The mean is the average of a distribution of values that are grouped symmetrically around the central tendency.

The median is the middle value of a sample. It is used, particularly, where the values in a sample are asymmetrically distributed around the central tendency.

The mode is the interval, in a frequency distribution of values, that contains more values than any other.

In a symmetrically distributed population, the mean, median and mode are the same.

The average value of a sample, on its own, is of only modest interest. Of equal (and often greater) relevance is the confidence we can place on the sample average as truly reflecting the average value of the population from which it has been drawn. This is most often expressed as a confidence interval, which describes the probability of a sample mean being a certain distance from the population mean. If, for example, the mean systolic blood pressure of 100 undergraduates is 124 mmHg, with a 95% confidence interval of ± 15 mmHg, we can be confident that if we replicated the study 100 times the value of the mean would be within the range 109–139 mmHg on 95 occasions. It is intuitively obvious that the larger the sample the smaller will be the size of the confidence interval.

Correlation

In clinical studies two, or more, independent variables may be measured in the same individuals in a sample population (e.g. weight and blood pressure). The degree of correlation between the two can be investigated by calculating the correlation coefficient (often abbreviated to ‘r’). The correlation coefficient measures the degree of association between the two variables and may range from 1 to −1:

If r = 1, there is complete and direct concordance between the two variables

If r = −1, there is complete but inverse concordance

If r = 0 there is no concordance.

Statistical tables are available to inform investigators as to the probability that r is due to chance. As in other areas of statistics, if the probability is less than 1 in 20 (p < 0.05) then by custom and practice it is regarded as statistically significant. There are, however, two caveats:

The 1 in 20 rule is a convention and does not exclude the possibility that a presumed association is due to chance

The fact that there is an association between two variables does not necessarily mean that it is causal. For example, a correlation between blood pressure and weight, with r = 0.75 and p < 0.05, does not mean that weight has a direct effect on blood pressure (or vice versa).

Correlation analyses can become complicated. The simplest (least squares regression analysis) presumes a straight-line relationship between the two variables. More complicated techniques can be used to estimate r where a non-linear relationship is presumed (or assumed); where the distributions deviate from normal; where the scales of one or both variables are intervals or ranks; or where a correlation between three or more variables is sought.

Expressions of benefit and harm

There are three ways in which the outcomes, in clinical studies, are expressed:

1. Binary (dichotomous) outcomes

2. Survival analyses

3. Continuous outcomes.

Binary outcomes are often used in the design and analysis of RCTs. Such outcomes are dichotomous (such as alive or dead). The results are usually expressed as the relative risk (or risk ratio – RR). In a trial where the outcome is (say) mortality, the relative risk is the ratio of the proportion of treated patients dying to the proportion of control patients dying. Usually, RR of <1 is suggestive of benefit; an RR of >1 is suggestive of harm. RRs are almost invariably reported with their 95% confidence intervals. If the boundaries of the 95% confidence intervals do not cross unity the results are generally statistically significant (at least at the 5% level).

Survival analyses. In studies in which individuals are observed over a long(ish) period of time, and in which it is unreasonable (or erroneous) to assume that event rates are constant, the technique of survival analysis is used. This is most commonly reported as the hazard ratio (HR) and its 95% confidence interval. The HR is the probability that, if an event in question has not already occurred, it will happen in the next (short) time interval. It has, broadly, a comparable interpretation to the RR.

Continuous outcomes. Studies such as that in Table 17.10 may report outcomes using one or more continuous scales. In this study of the effects of prednisolone in the treatment of Bell’s palsy, the House–Brackmann measure of facial nerve function was used as the outcome measure. Conventional tests of statistical significance using Student’s t-test, for example, can be calculated to assess whether the null hypothesis should be rejected.

Number needed to treat (NNT). As discussed earlier, the NNT is an estimate of the number of patients that need to be treated for one to benefit compared to no treatment. If the probabilities of the end-points with the active drug and no treatment (i.e. placebo) are respectively p_active and p_{no treatment} then the NNT can be calculated thus:

NNT = 1/ (p_active – p_{no treatment})

An analogous measure – the number needed to harm (NNH) – is the number of patients that need to be treated with a drug to cause one patient to be subject to a specific harm.

Only gold members can continue reading. Log In or Register to continue