Perspectives The case-cohort design in outbreak investigations 11

We describe the second case in Europe of verified treatment failure of pharyngeal gonorrhoea, caused by an internationally occurring multidrug-resistant gonococcal clone, with recommended first-line ceftriaxone 250 mg in Slovenia. This is of grave concern since ceftriaxone is last remaining option for empirical treatment. Increased awareness of ceftriaxone failures, more frequent test-of-cure, strict adherence to regularly updated treatment guidelines, and thorough verification/falsification of suspected treatment failures are essential globally. New effective treatment options are imperative.


Background
Neisseria gonorrhoeae has developed resistance to all antimicrobial drugs previously used as first-line treatment for gonorrhoea [1].Resistance to currently recommended first-line third-generation cephalosporins -cefixime and ceftriaxone -is emerging [1][2][3], and treatment failures with cefixime have been verified in Japan [4] and several European countries, namely Norway [5], the United Kingdom [6], Austria [7] and France [8].One failure to treat pharyngeal gonorrhoea with ceftriaxone, the last remaining option for empiric treatment, has also been verified in Europe (Sweden) [9].It is likely that treatment failures with ceftriaxone will initially accumulate for pharyngeal gonorrhoea because these infections are harder to treat than urogenital infections [1,10,11].It is of grave concern that during the past year, the first three extensively drugresistant (XDR) [1] N. gonorrhoeae strains that also had high-level ceftriaxone resistance were reported from Japan, France and Spain [8,12,13].
In this emergent situation of fear that gonorrhoea may become untreatable [1,8,12], the European Centre for Disease Prevention and Control (ECDC) has prepared a response plan for the European Union [14].The World Health Organization (WHO) has published the 'Global Action Plan to Control the Spread and Impact of Antimicrobial Resistance in Neisseria gonorrhoeae' [15].This report describes a ceftriaxone treatment failure of pharyngeal gonorrhoea in Slovenia in 2011, which is the second one strictly verified in Europe (and possibly globally).

Case description
In early September 2011, a Slovenian bisexual woman in her early 30s visited a dermatovenereologist in Ljubljana, Slovenia (Day 1).She had no symptoms of gonorrhoea, however, she was sampled and administered the internationally recommended first-line treatment of 1×250 mg ceftriaxone intramuscularly (Table ), based on the fact that she had had unprotected oral and vaginal sex with gonorrhoea-positive casual male partner in late August 2011 in Belgrade, Serbia.The partner could later not be traced in Serbia.
Microscopy of Gram-stained smear of a cervical specimen was negative for N. gonorrhoeae.However, two days later (Day 3), a pharyngeal culture was shown to be positive for N. gonorrhoeae, while the cervical culture was negative.Chlamydia trachomatis DNA was identified in an additional cervical sample, using the COBAS TaqMan CT Test v2.0 (Roche Diagnostics).During a follow-up visit seven days after the initial visit (Day 8), a test-of-cure (TOC) pharyngeal culture was taken and examination showed no signs or symptoms of pharyngeal gonorrhoea, and she was given doxycycline at a dosage of 100 mg twice a day, for seven days, for a concomitant chlamydial infection.However, two days later (Day 10) the TOC culture confirmed gonococci in a pharyngeal sample.About three weeks later (Day 30), the patient returned with symptoms of acute pharyngitis (pain, inflammation and fever) and was given one dose of 250 mg ceftriaxone intramuscularly and one oral dose of 1 g azithromycin.Finally, a follow-up examination after about four months (Day 173) showed no signs of infection, and a pharyngeal TOC culture was negative for N. gonorrhoeae (Table).The patient repeatedly reassured that she had not had any sexual contacts between the ceftriaxone therapy and the TOC.

Discussion
This study describes the second verified case in Europe (possibly globally) of treatment failure of pharyngeal gonorrhoea with the internationally recommended firstline treatment of 250 mg ceftriaxone, the last remaining treatment option.The failure was strictly verified in accordance with WHO recommendations [1,15], i.e. detailed clinical records were obtained, reinfection was excluded as much as possible, pre-and posttreatment isolates were indistinguishable using highly discriminatory typing, ceftriaxone MICs were elevated, and the isolates contained well-known cephalosporin resistance determinants.The reporting of the case was unfortunately delayed because it took several months before the patient returned for follow-up examination and TOC after the third antimicrobial treatment (to prove successful eradication of infections).
This case shows that ceftriaxone at a dosage of 1×250 mg may in rare cases not be enough for treatment of pharyngeal gonorrhoea caused by gonococcal strains with ceftriaxone MICs of 0.125 mg/L.A 250 mg ceftriaxone dose also results in median times of free ceftriaxone above the MIC of only 24.1 h (range: 10.5-52.2h) for the detected MIC of 0.125 mg/L [22], and rare treatment failures may happen in the lower range.Nevertheless, these cases are likely to be treatable with enhanced ceftriaxone doses or dual antimicrobial treatment that has already been introduced as first-line empiric treatment in the United States [10] and the United Kingdom [23].It may be crucial to promptly revise also other national and regional treatment guidelines, and a revision of the European guidelines from the International Union against Sexually Transmitted Infections (IUSTI) and WHO [2] are currently in progress.[12] and NG-MAST [16] of N. gonorrhoeae pre-and post-treatment isolates.b Treatment of concomitant C. trachomatis infection.
It is worrying that the gonococcus causing this treatment failure was assigned to MLST ST1901 and NG-MAST ST1407, which is a multidrug-resistant gonococcal clone that also shows decreased susceptibility and resistance to cefixime and is spreading worldwide [5,7,8,13,20,21,[24][25][26][27][28].The previously reported treatment failures with cefixime in Norway [5], Austria [7], France [8] and likely in the United Kingdom [6], were caused by this gonococcal clone or its evolving subtypes.This clone has also shown its capacity to develop high-level resistance to ceftriaxone [8,13].
In conclusion, the second case in Europe (possibly worldwide) of clinical failure using standard ceftriaxone treatment for pharyngeal gonorrhoea, caused by an internationally occurring multidrug-resistant gonococcal clone, has been strictly verified in Slovenia.An increased awareness of treatment failures with ceftriaxone, more frequent TOC (all cases of pharyngeal cases may be crucial), strict adherence to appropriate treatment guidelines, which need to be regularly updated based on antimicrobial resistance surveillance data, and thorough verification/falsification of suspected treatment failures (including subsequent tracing of sexual contacts of the index case with the treatment failure) are essential globally.A stronger focus on pharyngeal gonorrhoea, including increased sampling of pharyngeal specimens and promotion of condom use also when practising oral sex, is also crucial because pharyngeal infection is harder to treat than urogenital infection, relatively common, and is frequently an asymptomatic reservoir for infection and emergence of resistances [1,5].Ultimately, new options for effective treatment of gonorrhoea are imperative.

Surveillance and outbreak reports
Salmonella Paratyphi B var Java infections associated with exposure to turtles in Bizkaia, Spain, September 2010 to October 2011 Introduction Salmonella infections are predominantly acquired through the consumption of contaminated food, but contact with animals may also be an important source of infection [1].Reptiles are frequent carriers of Salmonella in their intestinal tract [2], they usually show no signs of illness and shed the bacteria in their faeces, contaminating the water and any surface in contact with them [3][4][5][6].
Although other reptiles can also carry Salmonella, turtles pose a special risk, as they are commonly kept as pets for children.S. Paratyphi B infections can cause enteric fever (paratyphoid fever) or gastroenteritis.In some cases, serious complications can occur (septicaemia, meningitis), especially in young children and immunocompromised patients [7].
S. Paratyphi B var Java shares the same somatic and flagellar antigens as S. Paratyphi B, but uses d-tartrate as a carbon source.This variant appears to be less virulent, causing infections characterised by watery diarrhoea, abdominal pain and fever, although infection can also be invasive.In sporadic cases and outbreaks, infection with S. Java has been associated with consumption of contaminated food, including salads, goat's milk cheese and poultry and with contact with reptiles and tropical fish aquariums [8][9][10][11].
The Epidemiology Unit of the Department of Public Health in Bizkaia (a territory of the Basque Country, in the north of Spain, with a population of nearly 2,150,000 inhabitants) identified, between September 2010 and October 2011, 14 cases of S. Paratyphi B infection (incidence rate: 0.65/100,000 inhabitants).In Spain, the most common Salmonella serotypes are Enteritidis and Typhimurium.S. Paratyphi B biovar Java represented 2.1%, 1.4% and 1.7% of the Salmonella strains isolated from humans and serotyped at the National Reference Laboratory for Salmonella in 2009, 2010 and 2011 respectively.As S. Java is an unusual serotype, an investigation was initiated to identify the risk factors.

Methods
A case was defined as a patient, resident in Bizkaia, who had an isolate of S. Paratyphi B var Java between September 2010 and October 2011.
Adult cases and the parents of the affected children were contacted by telephone and questioned using a standard questionnaire about potential risk factors, such as other cases of gastroenteritis in their environment, travel, consumption of suspected food items and animal exposure.Where contact with turtles was reported, a water sample was collected from the turtle's aquarium or terrarium for Salmonella testing.Another water sample was taken from the turtle tank at the shop where one of the turtles was bought, for laboratory analysis.The detection of Salmonella in the water samples was performed using enzyme-linked fluorescence assay (ELFA) method (bioMérieux´s VIDAS) and by culture (ISO 19250 Water quality-detection of Salmonella spp.).
Isolates from patients and environmental samples which were positive for S. Paratyphi B were submitted for confirmation to the reference laboratory, National Centre for Microbiology, Carlos III Institute of Health, Madrid, Spain.The strains were typed using phenotypic (lead acetate method) and molecular methods to detect the tartrate reaction [12].Susceptibility to antimicrobials was tested by the disc diffusion method according to Clinical and Laboratory Standards Institute (CLSI) guidelines [13].The panel included the following antimicrobials: ampicillin, cefalotin, cefotaxime, amoxicillin/clavulanic acid, tetracycline, streptomycin, kanamycin, gentamicin, nalidixic acid, ciprofloxacin, chloramphenicol, trimethoprim/sulfamethoxazole and a sulphonamide compound (sulfadiazine, sulfathiazole and sulfamerazine sodium).The Xbal-PFGE patterns of strains were compared according to the PulseNet protocol [14].
Three of the 14 cases, identified as S. Paratyphi B, were excluded from the investigation because they were not S. Java or its variant.

Results
Out of the 14 strains of S. Paratyhi B studied, eight were identified as S Paratyphi B variant Java (S.Java), three as possible monophasic variants of S. Java (S. 4,5,12:b:-), and three as S. Paratyphi B sensu stricto.The last three, which came from a family outbreak involving three siblings, produced different clinical manifestations, and were excluded from this description.
The 11 patients from whom S. Java or its possible monophasic variant was isolated were not related to each other, and developed a mild disease, with symptoms of gastroenteritis.
Except for two adults in their mid-20s and early 60s, all cases were children aged between three months and  a These samples were taken with a delay of five to 13 months after the infection.b Adults in their mid-20s and early 60s.
10 years.Six of the cases among children were male and three were female).
During the interviews, the only common factor found to constitute a risk according to the literature was having been in contact with aquatic turtles during the days before illness onset in six of the nine children, either at home (four cases), or at a relative's house (one case) or at school (one case).
The laboratory results show three different PFGE profiles, which we call type 1, 2 and 3 (Table , Figure 2).All strains were fully susceptible to all antimicrobials tested.
Three of the six samples of turtle's water yielded Salmonella Java, with the same PFGE patterns as the bacteria isolated from the children who had contact with them.Two of them were type 2, and the other was type 1.The three negative results came from samples collected more than five months after the infection.
The turtles were purchased at different shops and the supplier or suppliers could not be identified.
The PFGE patterns of isolates of patients with and without turtle exposure were indistinguishable, although the source of infection could not be found.All PFGE profiles were compared with those deposited at the PulseNet network and no match was found.
The water sample taken from the shop where the turtle of case 8 had been bought yielded Salmonella serogroup C.This turtle belonged to the subspecies Trachemys scripta scripta.The species of the other turtles are not known.

Discussion
Although we lacked a control group, the epidemiological and laboratory findings from our investigation indicate that turtles were the most likely source of infection with S. Paratyphi B var Java or its possible monophasic variant in this cluster of cases.Although any Salmonella serotype may be carried and transmitted by turtles, S. Java has been particularly associated with these reptiles [4].
For the first time, a possible monophasic variant of S. Java associated with reptile contact is described.This is the second time we find an association between contact with turtles and Salmonella infection.In 2008, following an increase in S. Typhimurium infections in our region, a case-control study was performed, which estimated the odds of infection to be 1.62 times higher if the case had been exposed to turtles (95% confidence interval (CI): 0.68-3.89).In this study, 67/145 (46.2%) of cases were children aged between one and four years and 24/138 (17.4%) of cases reported contact with turtles.The association between reptile exposure and Salmonella infection has been described in several countries [2][3][4][5][6][7][15][16][17][18].
Most cases of turtle-associated salmonellosis occur in young children, who are in the most susceptible age spectrum, probably because they usually have a closer contact with these pets, and play with the aquarium water, which is a good medium for the growth of Salmonella.Moreover, their hygiene practices tend to be worse than those of adults [2,15].In addition, parents are often not aware of the risk of infection associated with the presence of turtles in the household.Not all the cases in this cluster reported exposure to turtles.However, direct contact is not necessary for infection; environmental contamination and symptomatic or asymptomatic patients represent possible sources of infection that may have gone unnoticed.As Salmonella bacteria survive in the environment for a long time [2,5], indirect transmission can play an important role.Three of the six samples of turtle's water tested negative.However, Salmonella shedding can be intermittent and increase in response to stress like crowding, living in an environment with inadequate temperature, humidity or cleanliness, transportation, a change of habitat or excessive handling.A negative result doesn't rule out the possibility of intermittent water contamination [2,5].For this same reason, a mixed infection in the water of the shop where Salmonella serogroup C was found is possible.
In the United States of America (USA), the association between contact with small turtles and Salmonella infection lead, in 1975, to a ban on the sale and distribution of turtles under 10.2 cm in carapace length, except for scientific or educational purposes.As a consequence, an important reduction in the number of Salmonella infections was observed in the following years, especially among children [2][3][4][5].Since then, many sporadic turtle-associated salmonellosis cases have been detected.
In recent years, there has been an increase in the number of reptiles kept as pets, as well as in the number of infections linked to contact with reptiles, including more common serotypes, such as Typhimurium [2][3][4].Currently, an estimated 6% of Salmonella infections in the USA are caused by direct or indirect contact with reptiles [4].In February 2012, the Centers for Disease Control and Prevention (CDC) reported 132 cases of S. Paratyphi B var. L (+) tartrate + infection between 5 August 2010 and 26 September 2011.The median age of the patients in this outbreak was six years and of the 56 patients interviewed, 36 reported turtle exposure [19].
In Europe, Salmonella infection cases attributed to direct or indirect contact with reptiles have also been described, although the number is likely to be underestimated, as in many cases the source of infection is unknown [17].In Sweden for instance, between 1990 and 2000, 339 reptile-associated Salmonella infections were reported, accounting for approximately 5% of all reported cases [5].In this country, from 1970 to 1994, a certificate was required for the import of reptiles, stating that the animals were free of Salmonella, and the commercial distribution of turtles with a carapace length less than 10.2 cm was banned.When import regulations ceased, an increase in the number of cases was observed between 1996 and 1997.After a public education campaign launched in 1997, the number of cases decreased again [20].
Attempts to eliminate Salmonella from turtles by antibiotic treatment have not been successful, as the animals readily become reinfected from the environment, food or other turtles and can result in the development of antibiotic resistance.As Salmonella shedding may be intermittent and related to stress, it is difficult to determine whether turtles are free of bacteria [2].For this reason, the way to prevent transmission is to avoid contact of susceptible persons with turtles and to follow strict hygiene practices to minimise the risk of infection.
In the US, apart from the restrictions on the sale of small turtles, there are recommendations published by CDC for preventing reptile-associated salmonellosis, which include washing hands after handling reptiles and keeping reptiles away from food and food preparation areas [21].

Conclusions and recommendations
In conclusion, there is a risk of Salmonella infection linked to contact with turtles, which emphasises the need to give recommendations regarding ownership and handling of aquatic turtles and other reptiles kept as pets by young children.These recommendations can also apply to immunocompromised persons.It would also be appropriate to give information to potential buyers at points of sale about the risk of Salmonella infection and measures that can be taken to minimise this risk.
A report of this outbreak with the following recommendations was sent to the public health authorities and the Department of Agriculture in Bizkaia so that preventive measures can be taken.
Recommendations given for preventing Salmonella infection from turtles included: • washing hands with water and soap immediately after handling turtles (or other reptiles); • cleaning and disinfecting surfaces that have been in contact with the animal; • not using the kitchen to wash the aquarium/terrarium (if the bathroom is used, this should be disinfected after use); • avoiding contact of the turtle with food (turtles should not live in the kitchen or roam freely in the house); • avoiding contact of especially susceptible people (children under five years, pregnant women, patients with cancer or undergoing chemotherapy treatment, transplanted patients, persons with diabetes, hepatic conditions or other immunocompromised persons) with turtles and any object that has been in contact with them.

Introduction
Outbreaks are defined as any excess in the number of cases of disease that would normally be expected in a particular geographic area over a particular period of time [1].Outbreak investigations differ from standard epidemiological research as they are often conducted under time and resource constraints.Design options mostly depend on the outbreak setting, the size of the outbreak and of the population affected, whether or not the affected population is well defined and its members identifiable, and what measure of association is desired.In addition, the approach may vary depending on the pathogen (or the environmental hazard) and its mode of transmission, as well as time, staff and resource constraints.
The two main study designs generally considered by field epidemiologists in the investigation of outbreaks are the retrospective cohort design and the traditional case-control design.In the retrospective cohort, all members of a defined cohort are included in the study and information on their exposure to different factors is investigated retrospectively [2].Risk of illness in exposed and unexposed individuals is obtained and the measure of association is the risk ratio (RR).
Traditional case-control designs (also called 'cumulative' or 'classic' case-control) offer an efficient alternative when the source population (i.e. the population from which cases arose) is large and/or the outcome rare.Exposures in cases are compared to exposures in a sample of the non-cases (i.e. the controls) drawn from the same at-risk population, and the most common measure of association is the odds ratio (OR).
The case-cohort design is an alternative to the traditional case-control design.In the case-cohort design controls are randomly sampled from the source population, regardless of their disease status.
Although the case-cohort design has gained popularity in large prospective studies [3], its use in outbreak investigations has been limited [4,5].There is, to the best of our knowledge, no publication that explains, summarises and discusses the use of case-cohort designs in the context of outbreak investigations.
In that context, the aim of this paper is therefore to summarise the theory of the case-cohort design, illustrate its use in four different outbreak scenarios and discuss its strengths and limitations.

Description of the case-cohort design
The foundation of case-cohort design is generally attributed to Prentice who, in 1986, described it as an efficient alternative to a full cohort design in the context of prospective research when the collection and follow-up of covariate information in each cohort member is costly and time-consuming [6].Similar approaches were suggested by others under the terminology 'hybrid epidemiologic design', 'case-base' design or 'inclusive' case-control design [7][8][9].

Sampling and sample size
In a case-cohort design, all cases (or a random sample of all cases) and a random sample of the source population (i.e. the controls) are included in the study.
The controls may therefore include some of the cases included in the case group [10].
Figure 1 illustrates the sampling of cases and controls in the case-cohort study design, and compares this with the traditional case-control and retrospective cohort designs.
The sampling strategies for controls include the whole range of probabilistic sampling methods used in crosssectional studies, also including complex sampling designs.There are not many examples, but one is in a case-cohort study during and outbreak in Darfur, Sudan that used complex sampling to recruit controls [4].
Generally a little less statistical power is achieved with a case-cohort study, compared to a traditional casecontrol study, if both have an equal number of controls, inversely proportional to the primary attack rate (AR).A simple way of estimating the number of controls required for a defined power is to apply sample size calculations used in traditional case-control studies and multiply the number of controls by a weighting factor corresponding to the inverse of the proportion of non-cases in the initial cohort.For example, if the AR is 33% then 50% more controls (as (1-0.33) - =1.5) will have to be selected than in a traditional casecontrol study, whereas if the AR is only 5% the number of controls will only need to be increased by 5% (as (1-0.05) - =1.05).In some situations however, the AR will not be known at the start of the investigation.

Measure of association and analysis
Provided that cases are a random sample of all cases and the controls are sampled randomly from the source population, the cross product of exposed and unexposed cases and controls will yield a true estimate of the crude RR (allowing for sampling error), unlike the traditional case-control study where the OR obtained from the cross product of exposed and unexposed cases and controls will generally overestimate the RR (if true RR>1) or underestimate the true RR (if true RR<1).This inflation -or deflation -of the OR in case-control studies increases as the AR increases -or decreases -and also depends on the magnitude of the true RR (Figure 2).
Standard logistic regression can be used for multivariable analysis, in the same way as in a traditional case-control study, to obtain direct estimates of the adjusted RRs from the model output.This approach, taken in previous case-cohort studies [4,5], is limited however by the lack of precision around the estimates,

Figure 1
Comparing three study designs: case-cohort, case-control and retrospective cohort with standard errors generally being equal or larger than the true standard errors [10].This may not be a major constraint when a strong association is found for a particular exposure variable; however in situations where weak evidence of an association is found, this should be taken into account in the interpretation of the results.Several solutions have been proposed to deal with this [8,11,12].Schouten et al. [11] developed pseudo-likelihood risk models using logistic regression with a so called 'sandwich estimator' (or robust variance estimator) derived from the covariate matrix of the model output.Logistic regression is applied in traditional case-control studies, but RRs are obtained directly from the model output.The sandwich estimator adjusts the standard errors of the RR.This approach only requires common statistical software but may be more challenging if software commands are not readily available.

Outbreak scenarios
We will illustrate the case-cohort design through four commonly encountered outbreak scenarios, and discuss its strengths and limitations compared with traditional case-control and retrospective cohort designs.
Examples are either based on published outbreak investigations, or, if no such outbreak investigation was published, are fictitious for illustrational purposes.We chose examples which illustrate the design well and cover different types of scenarios where the case-cohort design might be considered.

Scenario 1: A point-source outbreak in a closed setting
Outbreaks occurring in closed settings, such as schools, cruise ships or parties are common.To illustrate design options in this context, let us imagine a Salmonella outbreak following a party attended by 400 people, of whom 100 developed symptoms of diarrhoea within two days following the event and were defined as primary cases (AR: 25%).
If contact details of all participants can be obtained, the first choice would be to conduct a retrospective cohort design.Let us assume that in a retrospective cohort study a particular food item (food x) emerged as the most important risk factor in a univariable analysis, with 80 ill with diarrhoea amongst the 140 exposed (AR: 57.1%) and 20 ill amongst the 260 unexposed (AR: 7.4%), giving a crude risk ratio of 7.4 (95% confidence interval (CI): 4.8-11.4).
Under time and resource constraints of outbreak investigations, there is often a need to collect data on smaller sample sizes, and the use of traditional casecontrol or case-cohort studies (in this case, nested in the cohort) could be envisaged.
The Table compares the results of the univariable analysis for food x obtained with a retrospective cohort to those obtained in a traditional case-control study and in a case-cohort study, in which the sample size would be half that of the cohort.The true RR is obtained in the case-cohort study, whereas, in the traditional casecontrol studies, the OR does not approximate the true RR because the overall primary AR is high (25%), as shown in Figure 2.
Table .Comparing measures of association in the retrospective cohort, case-cohort and traditional casecontrol studies Attributable risk fractions (the proportion of cases explained by the association=(RR-1)/RR) can also be calculated easily with case-cohort studies.
Arguably, in most outbreaks such as food-borne outbreaks the exact quantification of the risk increase associated with a particular factor may be unimportant

Figure 2
The relationship between odds ratio and risk ratio, for increasing attack rates AR: attack rate; OR: odds ratio; RR: risk ratio as long as there is evidence that that particular factor is associated with an increased risk, and over-or underestimating the RR may not matter that much.

Scenario 2: A vaccine effectiveness study during an outbreak
Outbreaks provide good opportunities to measure vaccine effectiveness (VE).Traditional case-control studies are often conducted when the source population is too large to conduct a retrospective cohort design [13].
In these studies VE is calculated as 1-OR of being vaccinated, where OR is assumed to approximate the RR.However, in studies of VE it is important to accurately obtain a precise estimate of the true RR.The use of the traditional case-control study in that context should therefore be discouraged given the difficulties in interpreting the OR [14].
The case-cohort design offers a suitable alternative in the context of VE studies during outbreaks, as it allows (i) to obtain true estimates of the RR and (ii) to randomly sample controls from the population without the need to enquire about disease history.
During a mumps outbreak in Switzerland, Richard et al. [15] investigated and compared the VE of two mumps vaccines.Cases were obtained from outbreak and surveillance data and controls were selected from a random systematic sample of GP registers, regardless of children's disease status.Similarly, Carrat et al. [16] used a case-cohort design to investigate influenza VE.Vaccination status in cases of confirmed influenza was compared to the vaccination status in controls randomly selected from GP registers, irrespective of whether or not they suffered from ILI during the influenza epidemic period.The design was particularly useful to obtain true estimates of the RR, and thus the VE, given the high incidence rate of ILI and influenza in the population.

Scenario 3: A food-borne outbreak at a restaurant
Food-borne outbreaks linked to restaurants are common.The use of a retrospective cohort design in restaurant outbreaks is often limited by the lack of identifiable controls, either because the guests' details have not been recorded or because the restaurant management may refuse to release details on their customers [17].Traditional case-control studies are therefore often seen as the only available option, in which controls are a convenient sample selected from the non-ill meal companions of cases [17][18][19].There may be few of these unaffected individuals, or they may not represent the average meal consumption of the customers as they tend to be more similar to the cases with regard to their meal consumption.In a situation where nonill meal companions were scarce, Giraudon et al. [19] instead used a case-case approach in their investigation of a Salmonella PT1 outbreak linked to a fast-food restaurant in London.They compared consumption in mild cases to that reported by severe cases assuming an exposure dose-response effect.
We suggest that in food-borne outbreaks linked to restaurants, where no customers' list is available, a case-cohort design could be performed, in which meal consumption in the cohort of customers (e.g. based on receipts or any other type of restaurant record) would be compared to meal consumption in cases.Limitations with this approach include the lack of adjustment for the possible confounding effects age and sex, and the assumption that all food and drinks served were consumed.Its advantage is a rapid test of hypotheses, with no need of selection and interviewing controls.This can be particularly useful during ongoing outbreaks where speed is crucial.

Scenario 4: Investigating multiple outcomes
The opportunity to study multiple outcomes is particularly helpful in outbreak situations because, unlike in standard epidemiological research, case definitions are often dynamic.Generally, the case definition is initially broad (sensitive) and is narrowed down (more specific) as more information is gathered (e.g.laboratory confirmation).
With case-cohort studies, hypotheses can be tested with different sets of cases (e.g. from the most sensitive to the most specific case definition) using only one sample of controls.
Moreover, in situations where several outbreaks occur at the same time, especially outbreaks linked to similar risk factors, the case-cohort design allows for one single control group to be used as reference group to investigate multiple outcomes.
For example, Martin et al. [5] used a case-cohort study to investigate a Campylobacter outbreak in the municipality of Söderhamn, Sweden, linked to the consumption of communal water.Although the number of confirmed campylobacteriosis cases was small (n=101) in comparison to the population of Söderhamn (n=27,765), the use of a traditional case-control study was complicated by the fact that another large outbreak of acute gastrointestinal illness (initially thought to affect more than 20% of the residents) occurred simultaneously, possibly including some unconfirmed cases of campylobacteriosis and possibly linked to the same source.A case-cohort study was conducted, and the control group was a simple random sample of the community, thus including some individuals with gastrointestinal illness.The investigation found that consuming communal water increased the risk of both campylobacteriosis and acute gastrointestinal illness, and the risk increased with the amount of water consumed.

Conclusions
We have described the use of the case-cohort design in field epidemiology, and illustrated its strengths and weaknesses through examples.
Among the advantages we identified is that a true estimate of the RR is possible.Although the OR may be good enough in most outbreak situations, there are situations (in particular VE studies) where obtaining a precise estimate of the true RR is important.
Further, the control group represents a random sample of the source population, and detailed disease history is therefore not required.This is particularly advantageous when cases and controls are sampled from different source databases, for instance a surveillance database for cases and a GP practice register for controls.
In addition, the control group can easily be used as a reference group to investigate multiple outcomes.
There are also a few limitations such as reduced statistical power compared with a traditional case-control study and the few analytical challenges, which can be addressed, but need more statistical expertise than a traditional case-control design.

Introduction
Chlamydia trachomatis endocervical/urethral infection, caused by serotypes D to K is the most common bacterial, treatable sexually transmitted infection worldwide [1,2].As up to 80% of cases are asymptomatic, C. trachomatis can be spread unknowingly and remains largely undiagnosed [1,2].The prevalence of the infection in Europe varies according to the population, setting, country, resource allocation for surveillance and prevention and national reporting system, if there is one.A systematic review of C. trachomatis infection among asymptomatic unscreened European women showed that the prevalence ranged from 1.7% (among women aged 15-40 years in the United Kingdom in the mid-1990s) to 17% (among women aged 15-55 years in France in the late 1980s) and was more than 5% in the majority of the countries examined [3,4].More recently the European Centre for Disease Prevention and Control (ECDC) described surveys from seven countries, estimating a population prevalence of 1.4-3.0% in people aged 18-44 years [5].They also reported that overall trends over time across Europe appeared to be increasing, from 1990 to 2009, although data were not available from Bulgaria, Czech Republic, France, Germany, Italy, Liechtenstein and Portugal [6].Moreover, the organisation of the control of C. trachomatis infection varied widely, with many countries having no organised activities until 2009 [7].
Since treating complications is costly in both psychosocial and financial terms, and is often unsuccessful [15], screening is critical for the early detection and treatment of uncomplicated C. trachomatis infection, the control of the overall prevalence of the infection in the population and thus the reduction of transmission and finally for the reduction of treatment costs.
C. trachomatis screening programmes exist in only two European countries (England and the Netherlands) and in the United States: they are opportunistic or pro-active and are mostly directed at young women aged under 25 years [7,16].Sweden, although lacking nationally organised screening programmes, is the first country in the world to offer testing for C. trachomatis infection, treatment and partner notification -all free of charge -throughout the country.It is also the first to have a national diagnostic and reporting system [5].In these four countries, after substantial decreases in complication rates of C. trachomatis infection at the end of the 1980s and early 1990s, further decreases in pelvic inflammatory disease and ectopic pregnancy rates after 2000 were observed [7,[16][17][18][19][20].
Unfortunately, reliable and recent data concerning C. trachomatis control in Italy are lacking, except for those in studies such as that of the Italian MEGIC Group (Multicentre Epidemiology Group for Investigation of Chlamydia trachomatis) that reported a prevalence of C. trachomatis infection of 3.9% among 1,321 asymptomatic women [21] or that of the STD Surveillance Working Group, which described 809 female incident cases from mainly dermatology and venereology departments and a few gynaecological departments between 1991 and 1996 [22].
There is no screening policy for C. trachomatis infection in Italy.A national women's health report released in 2008 suggested for the first time that women should be tested for C. trachomatis when they have their first cervical smear test [23].In order to understand if a screening strategy would be appropriate, the prevalence of the infection needs to be ascertained and there needs to be a preliminary analysis of the epidemiological variables in the population at risk, as well as a surveillance network.No existing epidemiological model can be applied to a different population without analysis and adjustment.New, larger epidemiological analyses are therefore needed in Italy to plan specific and effective strategies for the surveillance and screening of C. trachomatis infection in the country.The purpose of this study was to investigate the prevalence of C. trachomatis endocervical infection and its determinants in a large population of sexually active women aged 15-55 years attending an outpatient service of a cervico-vaginal pathology unit in Rome over a 10-year period.

Patient population
Between January 2000 and December 2009, a total of 7,620 women (aged 13-58 years) attending the outpatient service of the Cervico-Vaginal Pathology Unit in the Department of Gynaecology and Obstetrics of Sapienza University in Rome were examined for genitourinary symptoms or routine gynaecological examination.
A team of gynaecologists collected socio-demographic and behavioural data, as well as clinical data, for each woman during this time, using our model of clinical record taking for sexually transmitted infections -a structured questionnaire.The data were archived as digital files.
The self-administered, structured, paper questionnaire comprised 25 questions on socio-demographic characteristics, sexual behaviour, reproductive history, and tobacco, alcohol and drug use.
Testing for C. trachomatis infection, along with testing for human papillomavirus (HPV) and N. gonorrhoeae infection and vaginal wet mount examination, was offered to all sexually active women presenting to the Unit.
Women who refused to be tested for C. trachomatis and/or to answer the questionnaire and/or were not sexually active were excluded from the study (n=651).
According to these criteria, a total of 6,969 sexually active women aged 15-55 years who were tested for cervical C. trachomatis infection were enrolled.The women were categorised as symptomatic if they presented with either dysuria or pelvic pain or both (symptoms typical of C. trachomatis infection).Women not exhibiting either of these symptoms were classified as asymptomatic.They were then further categorised according to whether they were seeking care for family planning, infertility routine gynaecological examination or matters related to pregnancy.
All participating women gave written informed consent.The research was carried out in compliance with the Declaration of Helsinki [24] and was approved by the local ethics committee (reference number 148/11, 2022).Data were stored and managed according to Italian privacy rules [25].

Examinations performed
On a scheduled visit, during the gynaecological examination, an unmoistened sterile speculum was inserted into vagina, so that vaginal walls, fornices and cervix could be evaluated for any erythema and colour and viscosity of any discharge.The pH of the vaginal walls was measured using colorimetric paper.For wet mount examinations, vaginal fluor samples were collected from lateral fornices by a wooden Ayre's spatula, mixed first with saline and then with 10% potassium hydroxide, on two different slides, and immediately observed under a phase contrast microscope [26].
A 'whiff test' using 10% potassium hydroxide was performed for each sample in order to detect abnormal amine production by anaerobes [27].
Wet mount examination allowed the vaginal microflora (predominance of lactobacillary morphotypes) to be assessed and Trichomonas vaginalis to be detected (in order to investigate coexisting sexually transmitted infections).In addition, we also looked for bacterial vaginosis-associated clue cells, aerobic vaginitis-associated pleomorphic bacteria, yeasts and white blood cells.
Samples were taken from the endocervix for detection of C. trachomatis and from the ecto-endocervix for detection of HPV DNA, as described below.

Chlamydia trachomatis
Endocervical swabs were tested for the presence of C. trachomatis using the BD ProbeTec ET System (Becton, Dickinson and Company, United States).These assays amplify C. trachomatis DNA in separate wells and monitor inhibition of amplification for each specimen using strand displacement amplification and detection by fluorescent energy transfer probes, producing a method-other-than-acceleration (MOTA) score for each specimen The original algorithm involved retesting specimens with MOTA scores between 2000 and 9999.

Human papillomavirus
DNA was extracted from cervical samples using QIAampTissue Kit (Qiagen, Italy) and then genotyped by sequencing a 450-base pair fragment amplified from the L1 region of HPV DNA [29].Sequence homology was determined using BLAST and ClustalW programs.

Neisseria gonorrhoeae
Identification of N. gonorrhoeae was carried out by growth on media selective for pathogenic Neisseria species (Oxoid) incubated for up to 48 hours in 5-10% CO 2 at 35-37 °C.Colonies obtained were identified by API NH (bioMérieux) [30].

Statistical analysis
The chi-square test was used to analyse contingency tables; the t-test was used to compare means and odds ratios (ORs), with 95% confidence intervals (CIs), in order to measure the strength of association between C. trachomatis infection and behavioural and clinical characteristics and age.
We used the Cochran-Armitage test to assess the possibility of a linear trend in the observed patterns for number of lifetime sexual partners and increasing age.
Statistical tests were considered significant if p was 0.05 or less.A stepwise backward logistic regression analysis, entering the variables significantly associated with C. trachomatis infection, was used to assess the effect of more than one variable at a time and to identify possible confounding factors in the range of test values under consideration.Statistical analysis was performed using SPSS version 18.0.

Results
A total of 366 (5.2%) of the 6,969 women sexually active women enrolled in the study tested positive for C. trachomatis endocervical infection (Table 1).
Prevalence of C. trachomatis infection by year is shown in the Figure : the p value for the chi-square statistic was not statistically significant (p=0.938)(the chi-square test for the resulting 2×10 contingency table tested the null hypothesis of no association against the alternative hypothesis of an association of some sort).Thus prevalence and time appeared not to be associated and were not expected to have a linear correlation over the study period.
A total of 4,620 (66%) of the women were asymptomatic for C. trachomatis infection: 256 (5.5%) of them tested positive.This prevalence was slightly higher than that in the 2,349 symptomatic women (4.7%), but the difference was not statistically significant (p=0.1289).Of the 366 women who were positive for C. trachomatis infection, 256 (70%) were asymptomatic.
Prevalence was also slightly higher among women without clinical signs of infection (238/4,328; 5.5% compared with those with signs (128/2,641; 4.8%), but this difference was also not statistically significant (p=0.2362).
Univariate analysis of sexual and reproductive history and of age (Tables 1 and 2) highlighted a significant association of C. trachomatis infection with age under 40 years, having never been pregnant, smoking, use of oral contraceptives and multiple lifetime sexual partners: women with two to four partners had a slightly higher risk of infection (in comparison with women who had had one partner); women with five to nine partners had double the risk; having had more than nine partners was linked to a threefold higher risk.The p value for the Cochran-Armitage test (p<0.0001)suggested an underlying positive linear trend between number of lifetime sexual partners and prevalence of infection.
Comparison of the prevalence of C. trachomatis infection in stratified age groups with that in women over 49 years of age showed that teenage women aged 15-19 years had the highest increased risk of infection (OR: 4.55 (95% CI: 1.90-10.89);p=0.0002) and that the odds ratios for the remaining strata declined with increasing age.The p value for the Cochran-Armitage test (p<0.0001)suggested an underlying negative linear trend between age and prevalence of infection.
Further univariate analysis showed that the prevalence of the infection was similar (no statistical significance) whatever the reason for seeking care (

Discussion
To the best of our knowledge, this is the first study reporting on the epidemiology of C. trachomatis infection in Italy in a large sample of a diverse group of women over a long period of time.The mean prevalence of the infection was high (5.2 %) and showed no linear trend over time.The prevalence in asymptomatic women was higher than that observed in 1990 by the MEGIC group (5.5% vs 3.9%, respectively) [21].
In symptomatic women and in those seeking care for infertility the prevalence in our study (4.7% and 4.9% respectively) was similar to that reported by the same group (5.0% and 5.4%, respectively) [21].These findings may reflect the lack of control and screening activities in Italy.
We also found a high prevalence of C. trachomatis infection in pregnant women (5.3%), i.e. those seeking obstetric care (  The overall chi-square statistic was 6.255 (the chi-square test for the resulting 2×10 contingency table tested the null hypothesis of no association against the alternative hypothesis of an association of some sort).The p value for the chi-square statistic (p=0.938) was not statistically significant.
than five), have also been highlighted by research groups worldwide in various populations [7,16,31].We found that the highest prevalence of infection (10.8%) was associated with a nearly fivefold increased risk of infection (as an independent factor, it showed a twofold increased risk) in women aged 15-19 years.
Before 2008, C. trachomatis control activities in Italy consisted of case management in dermatovenereology clinics with Chlamydia testing for symptomatic people only [7].C trachomatis testing is currently recommended for women at the time of their first cervical smear test, which takes place when women are 25 years of age in Italy.To the best of our knowledge, no report on the uptake and results of this testing recommendation is yet available.However, our data suggest that women aged under 25 years, and in particular those under 20 years, would be the core population of a good testing policy and a hypothetical C. trachomatis screening programme, as in other screening programmes worldwide [7,16].Thus, the current Italian policy could be ineffective.The high prevalence of infection observed until the age of 40 years -which is a novel aspect of our findings -could also lead to a more extensive testing strategy.Although being aged 25-39 years was not an independent risk factor for infection, our data suggest that older women should also be tested.
Furthermore, as prevalence in women with signs or symptoms of infection did not differ statistically from that in women with no signs or symptoms in this study, case management appears to be an insufficient Chlamydia control activity.
The prevalence of infection among women seeking care for family planning was also high (5.4%):despite the low number of women in our study who sought advice for family planning, given the high number of women who usually attend this type of service and their young age, we suggest that family planning clinics could be sentinel for Chlamydia surveillance or setting for Chlamydia opportunistic screening.
Our data also show that having HPV or T. vaginalis infection was associated with a fivefold higher risk of C. trachomatis coinfection, as expected in groups at higher risk as a result of age and behaviour [32,33].
In our logistic regression, HPV was not significantly associated with C. trachomatis infection, suggesting that age and multiple partners could be possible confounding factors, while T. vaginalis infection was an independent risk factor for C. trachomatis infection.It is possible that severe inflammation of the cervix due to T. vaginalis infection may make the cervix more susceptible to C. trachomatis infection.It could therefore be suggested that patients diagnosed with T. vaginalis infection should be tested for C. trachomatis or even given treatment for C. trachomatis infection without being tested, as proposed by Lo et al. [33].
Data on N. gonorrhoeae and C. trachomatis coinfection in Italy are limited, but our findings on N. gonorrhoeae seem to be consistent with those reported in 1998 by a dermatovenereology network, which found that fewer than 1% the infections in 44,438 individuals with sexually transmitted infections were N. gonorrhoeae cervical infections [22].
We also found a statistical association of C. trachomatis infection with absence of previous pregnancies, use of oral contraceptives and smoking.However, as they were not shown to be statistically associated with infection in the logistic regression final model, age, having multiple lifetime sexual partners and T. vaginalis infection are likely to be confounders, in contrast to the findings of others [34][35][36].
The lack of statistical association between C. trachomatis infection and condom use (as a protective factor) is unexpected, given the findings of others [21,37].This could be considered a result of incorrect condom use and lack of health education.It could also be that some of the women were not entirely truthful when providing details of the type of contraception they used.There are probably some methodological limitations in the epidemiological study of condom effectiveness in preventing C. trachomatis infection, as has been highlighted by Warner et al. [37].
A new C. trachomatis variant was detected in 2006 following an unexpected 25% decrease in the number of infections in a Swedish county [38,39].As we used the Becton Dickinson ProbeTec -which detects the new variant -the presence or absence of the variant in Italy has no impact on our prevalence data.However, as no data are available on the type and distribution of C. trachomatis diagnostic methods used in Italy, nor on whether this variant is present among Italian women, surveillance is also needed to provide such information.
In conclusion, the prevalence and determinants of C. trachomatis infection observed in this study seem to highlight the need for a focus on control activities in Italy, with special attention to standardisation of diagnostic tests and women aged under 25 years, who would be the core population of a screening programme.

Figure 1 a
Figure 1Cases of Salmonella Java and its possible monophasic variant by age group and exposure to turtles, Spain, September 2010-October 2011 (n=11)

Figure 2
Figure 2 PFGE profiles of cases, Salmonella Paratyphi B var Java infections, Spain, September 2010-October 2011

Table
Details of verified ceftriaxone treatment failure of one case of Neisseria gonorrhoeae pharyngeal infection, Slovenia, September 2011 b.i.d.: twice a day; CT: Chlamydia trachomatis; GC: Neisseria gonorrhoeae; IM: intramuscular administration; MIC: minimum inhibitory concentration; MLST: multilocus sequence typing; NA: not applicable; NG-MAST: Neisseria gonorrhoeae multi-antigen sequence typing; PCR: polymerase chain reaction; PO: per oral administration; STD: sexually transmitted diseases.a MIC (mg/L) as determined by Etest, MLST

Table
Comparing measures of association in the retrospective cohort, case-cohort and traditional case-control studies OR: odds ratio; RR: risk ratio.a Variance derived from a first-order Taylor series approximation.

Table 1
Univariate analysis of age and sexual and reproductive history of women tested for Chlamydia trachomatis infection, Cervico-Vaginal Pathology Unit, Sapienza University, Rome, Italy,2000-2009 (n=6,969) a Unless otherwise indicated.b Where relevant.The t-test compares the mean values for women who tested positive for C. trachomatis and those who were negative.c Not statistically significant.

Table 2
Univariate analysis of reasons for seeking care, clinical features, contraceptive use and smoker status of 6,969 women attending as outpatients the Cervico-Vaginal Pathology Unit, Sapienza University, Rome, Italy, 2000-2009 Figure Prevalence of Chlamydia trachomatis infection in women tested at the Cervico-Vaginal Pathology Unit, Sapienza University, Rome, Italy, 2000-2009 (n=6,969) a Not statistically significant.b Dysuria or pelvic pain.c Cervical erythema, inflammation or discharge.

Table 3
Univariate analysis of other sexually transmitted infections in 6,969 women attending as outpatients the Cervico-Vaginal Pathology Unit, Sapienza University, Rome,Italy, 2000Italy,  -2009 CI: confidence interval; HPV: human papillomavirus.a Women coinfected with T. vaginalis and HPV (n=41) are not included.