The “Shaken Baby” syndrome: pathology and mechanisms

Shaken Baby syndrome is generally, but not exclusively, diagnosed in infants under 1 year of age, the peak age being 10–16 weeks. Boys represent 65% of cases and are younger at presentation [153, 158]. The diagnosis is characterised by the triad of retinal haemorrhage (RH), thin-film bilateral or multifocal subdural haemorrhage (SDH) and encephalopathy. A mechanistic explanation and pathological description of the three components of the triad will be discussed in the context of our current understanding of the anatomy and physiology of the brain and its coverings in the first year of life. Not all babies presenting with the triad will die, and neuroradiology rather than neuropathology is the cornerstone of diagnosis in babies who survive. Interpretation of imaging depends on understanding the neuropathology and wherever possible reference will be made to the correlation of pathological with radiological appearances.

In the early 70s Guthkelch [63] and later Caffey [18] suggested that this triad could result from whiplash or shaking injury. Guthkelch, noting that not all babies with SDH had external marks of injury to the head, suggested that shaking rather than striking the infant might be the cause. Extrapolating from the results of contemporaneous biomechanical studies on adult primates, he suggested that the whiplash of shaking may shear bridging veins leading to bilateral thin-film subdural haemorrhage, which he noted was quite unlike the unilateral subdural bleeding typically described in adults. He wrote “Moreover, since one would expect that the child is often grasped more or less symmetrically by chest or limbs the rotation-acceleration strains on the brain would tend to occur fairly symmetrically also, in an anteroposterior direction. This may be the reason why infantile subdural haematoma is even more often bilateral—for example in 14 of 18 cases (78%) of my earlier series—than subdural haematoma in adults for which the proportion of bilateral cases does not exceed 50%”.

Duhaime studied both biomechanical and clinical aspects of Shaken Baby syndrome and wrote perhaps the most exhaustive studies and reviews of the condition in the late 80s. In a review of 57 patients with suspected shaking injury, all 13 who died had evidence of impact head trauma. Eight had skull fractures and bruises, five had externally visible scalp bruises and six had “contusions and lacerations” of the brain. Her biomechanical studies led her to conclude that the acceleration force generated by impact exceeded that caused by shaking by a factor of 50. The following year, Duhaime wrote “it is our opinion, based on the clinical data and the studies outlined, that the “Shaken Baby syndrome” is a misnomer, implying a mechanism of injury which does not account mechanically for the radiographic or pathological findings” [38].

Others have repeated these biomechanical studies and shown that an adult shaking a dummy cannot generate the forces considered necessary to produce subdural bleeding [31, 125]. In contrast, Roth, using finite element modelling of the material properties of bridging veins and the angular velocities measured by Prange [125], calculated that shaking could generate sufficient force to cause BV rupture. The peak force considered necessary to do so in this model was equivalent to that generated by a 1.25 feet fall [133].

Duhaime devised an algorithm for the diagnosis of non-accidental injury on which many subsequent studies have been based. It assumes that a short fall cannot explain the triad: “falls clearly described as less than 3 feet in height were designated as “trivial” trauma and when given as an explanation of a high-force injury, along with variability in the history or a developmentally incompatible scenario, non-accidental injury was presumed” [39]. Clinical and biomechanical studies have demonstrated the error in this assumption; there are many reports of babies suffering intracranial bleeding, sometimes fatal, after low falls [3, 55, 66, 168] and laboratory studies have shown that the forces generated by even a 25-cm fall are twice those generated by maximal shaking and impact onto a soft surface [147].

An exhaustive study of the published literature over a period of 32 years found only 54 cases of confessed shaking, of which only 11 had no evidence of impact and could be considered pure shaking. There are only three published reports of witnessed shaking; all three infants were already collapsed before the shaking event [91, 140].

Since the initial work of Guthkelch, the importance of “rotational forces” as the mechanism of intracranial injury has been emphasised. Many have mistakenly assumed that rotational forces require shaking. There is no doubt that rotation is a potent cause of intracranial injury, but virtually any impact to the head will also cause rotation because the head is hinged on the neck. Holbourn wrote in 1943 “rotations are of paramount importance” and “If the head is so well fixed that it cannot rotate at all when it receives a blow there will be no rotational injury” [70]. It is reasonable to assume that the infant with a weak neck would be even more vulnerable to hinging of the head on the neck than the adult with developed musculature and full head control. While rotational acceleration/deceleration is important in causing brain damage, there is absolutely no evidence that it requires shaking or swinging. While shaking does cause rotational forces, their magnitude is insufficient to cause intracranial injury; biomechanical studies have shown that impact and falls cause far greater rotational forces [31, 37, 125].

Neuropathological studies have had enormous implications for the shaking hypothesis. Geddes showed that brain swelling and HII were virtually universal in babies thought to have suffered non-accidental injury, but very few had traumatic axonal injuries. Where, the present axonal injury was at the craniocervical junction [57, 58]. The clinical implication is that the signs of encephalopathy are due to hypoxia and brain swelling. As the pace at which swelling occurs is variable, there is an opportunity for a “lucid interval”, in contrast to immediate concussion as expected from diffuse axonal injury. In the majority of infants with the triad, the brain damage is non-specific and unless there is injury at the craniocervical junction, the diagnosis of shaking can be no more than speculation.

In 2009 the American Academy of Paediatricians, followed in early 2011 by the UK Crown Prosecution Service, accepted that the term Shaken Baby syndrome should be dropped because it did not exclusively explain the triad of findings, although confessions supported the role of shaking. The term non-accidental head injury (NAHI) has since been widely adopted [27, 32].

While shaking is no longer a credible mechanism for NAHI, there remains no doubt that inflicted head injury does occur, but its clinical recognition remains problematic. “There is no diagnostic test for inflicted brain injury, the diagnosis is made on a balance of probability and after careful exclusion of other possible causes of the clinical presentation” [99] and “there is no absolute or gold standard by which to define NAHI” [69].

In arguing “The Case for Shaking” Dias [35] began with the statement “Unfortunately, nobody has yet marshalled a coherent and comprehensive argument in support of shaking as a causal mechanism for abusive head injury” and concluded “the consistent and repeated observation that confessed shaking results in stereotypical injuries that are so frequently encountered in AHT-and which are so extraordinarily rare following accidental/impact injuries-IS the evidentiary base for shaking”.

This very definite statement indicates that, 40 years after it was first proposed, the shaking hypothesis now rests upon confession evidence.

How reliable are confessions? Clinical evidence of impact is found in up to 63% of confessions of shaking only [35] and when imaging evidence is considered “No correlation was found between repetitive shaking and SDH densities.” [2]. This review will take a pragmatic approach, addressing the evidence provided by detailed examination of the tissue in babies who manifest the triad.

The foregoing discussion illustrates the considerable responsibility for the pathologist who may be presented with a clinical diagnosis based on dubious criteria, the reliability of which will depend on the extent to which other possible causes have been excluded.

The differential diagnosis of a baby with the triad is wide and includes birth difficulties, coagulopathy, arterial occlusive disease and venous thrombosis, metabolic and nutritional disorders, infections, hypoxia–ischaemia (e.g. airway, respiratory, cardiac, or circulatory compromise) and seizures. Multifactorial and secondary cascades are common, for example “trivial trauma” in the context of predisposing or complicating medical conditions such as prematurity, pre-existing subdural haemorrhage, coagulopathy and infectious or post-infectious condition (e.g. recent vaccination). Death occurring within the context of a recent vaccination should be reported to the appropriate agencies.

A complete and thorough review of current and past medical history involving scrutiny of perinatal and neonatal records, laboratory tests and the clinical management of the child is required. If the baby dies the gold standard is a thorough and complete autopsy where neuropathology has a key role.

Many alternative diagnoses may not have been considered or test results may not be available before pathological opinion is required. Once pathological conclusions have been reached, they can be assessed in the context of all the available information.

Neuropathologists may not have the advantage of attending the autopsy to see for themselves the evidence from other sites and particularly to be present when the skull is opened to identify bleeding and its potential sources. The brain may be received whole and fixed; if giving a second opinion, only blocks and slides may be submitted. Residual fixed brain slices should be requested; they frequently yield evidence that may not have been appreciated on the first examination. In my experience, cortical veins and focal congestion or thrombosis are often overlooked. The dura and spinal cord are essential parts of the examination.

Sampling

Standard representative blocks should be taken from all brain areas and all levels of the brainstem and spinal cord, with at least three blocks from each level of the brainstem and the cervical cord, to include nerve roots and dorsal root ganglia.

The dura must be carefully examined by naked eye as well as microscopically. Old, healing subdural membranes can be difficult to see with the naked eye as they form a thin, light brown and often uniform layer (Fig. 1b). The dural sinuses must be carefully examined and sampled. Intradural bleeding is the most common posteriorly, in the spongy tissue around the torcula, in the posterior falx and in the tentorium and these areas, as well as convexity dura, should be sampled.

Dura from the spinal cord is informative for two main reasons. First, unlike the cranial dura, it is not routinely stripped from the underlying arachnoid barrier membrane during autopsy. Foci of the normal in vivo apposition of these membranes, as well as minor bleeding into the subdural compartment may be seen in spinal dura (Fig. 2). Second, it is common for intracranial subdural blood to track into the spinal subdural compartment, and sometimes this blood is the only evidence of old subdural haemorrhage.

At the outset, it is important to appreciate the anatomical location of subdural bleeding, which, in fact, originates within the deep layers of the dura. The skull bones, periosteum and meninges develop by condensation from the same mesenchymal layer and the dura forms a single functional unit with the arachnoid barrier layer [98]. In life there is no subdural space; “the traditional concept of a virtual, slit-like subdural space is in error” [53, 65]. When the dura is stripped from the brain surgically, at autopsy and by bleeding, the loosely adherent cells of the dural border layer are torn apart and an artificial space is created. A careful examination of the outer surface of the arachnoid membrane will reveal cellular remnants of the dural border cell layer. Similarly, adherent arachnoid cells can sometimes be identified on the deep dural layers. (Fig. 2)

When a baby presents to hospital, it is often the radiological diagnosis of SDH that raises the question of NAHI and significantly influences subsequent management. The importance of this element of the triad places an onus on the pathologist to establish and describe the sources and nature of infant SDH.

The typical pattern of subdural haemorrhage in babies with the triad is of a bilateral thin film over the cerebral convexities and in the posterior interhemispheric fissure [22, 40, 189].

The distribution of subdural blood is not a reliable indication of its cause. Rather, radiological studies have demonstrated that the distribution is a function of age and that redistribution occurs by gravity and sedimentation [33, 47, 167]. MRI studies show spinal subdural haemorrhage in almost half of babies with intracranial subdural haemorrhage, sometimes in direct continuity with posterior fossa blood. The location in the most dependent spinal areas, dorsally at the thoracolumbar level, indicates gravitational redistribution [86]. The spinal dura extends beyond the dorsal root ganglia where it blends with the sheath of the nerve roots, allowing subdural blood to track out into them (see Fig. 10).

The bilateral widespread thin film distribution of infant SDH differs from the adult form, where subdural haemorrhage generally forms a unilateral localised mass within the convexity dura [63]. There are several potential explanations for this difference. First, the mechanism for SDH in an infant may be fundamentally different from that in an older child or an adult with a mature skull. Second, the infant dura is far less collagenised than the adult dura, and its fibroblasts are widely separated by a loose matrix (Fig. 3) allowing ready dispersion of blood. Finally, infant subdural haemorrhage is frequently not solid but “thin and easily tapped” [155, 167] allowing easy dispersion.

SDH does not need to be large or space occupying to cause clinical symptoms. Although the blood is physically separated from the surface of the brain by the arachnoid barrier layer, it causes cerebral irritation, and clinical manifestations may occur without obviously raised intracranial pressure.

The volume of blood seen on scans may be very small, 2–3 ml of blood are sufficient for its radiological identification; between 1 and 80 ml (median 10 ml) was found in babies with blunt force injuries to the head [111]. This poses a problem for the pathologist as this small volume of blood could readily be overlooked as the skull is opened and large vessels are cut, as they inevitably will be if special autopsy techniques are not employed. “At autopsy, the subdural hemorrhage may consist of only 2 to 3 ml of blood and may not be observed if the prosector does not personally inspect the subdural space as the calvarium is being removed. Extreme caution should be taken to not misinterpret as premortem subdural hemorrhage the blood draining from the dural sinuses when these are incised at autopsy” [22].

Needling the cisterna magna before opening the skull may identify extracerebral blood or fluid collections. In order to examine the bridging veins, the skull is opened by parasagittal cuts lateral to the superior sagittal sinus [82]. By carefully lifting the midline bony strip, the bridging veins can be visualised and any extracerebral blood and fluid identified before veins or sinuses are cut (Fig. 4). Some claimed to be able to establish the integrity of bridging veins by retrograde dye injection via the superior sagittal sinus [104, 154].

The four most important potential sources of subdural bleeding are the bridging veins (BV), the dura itself, the vascular membrane of a healing subdural haemorrhage, and a ruptured intracranial aneurysm.

Bridging veins

It is widely believed that subdural bleeding results from mechanical tearing of bridging veins [23]. However, it is extremely hard to find any pathological description of ruptured BV in infants with SDH. Cushing, describing his surgical and pathological observations in the newborn wrote “In two of the cases that I have examined I have satisfied myself that such ruptures were present. “A positive statement, however, cannot be given even for these cases, since the dissection and exposure, difficult enough under any circumstances, owing to the delicacy of the vessels, is the more so when they are obscured by extravasated blood”. We have not moved a long way on this issue in the last century. Voigt [171] described disruption of bridging veins in adults, but they were characterised by local subarachnoid bleeding, and he wrote “Most striking in these cases is the absence of a noteworthy subdural hematoma”. Duhaime [38] did not demonstrate BV rupture in her autopsied cases, but hypothesised that the point of bridging vein rupture is in the subarachnoid space, giving rise to both subarachnoid and subdural bleeding. Bell [9] illustrates a thrombosed bridging vein which she described as the site of traumatic rupture, but venous thrombosis is common in hypoxic and ventilated infants and does not provide robust evidence of BV rupture. Maxeiner [105] claimed to demonstrate ruptured BV using autopsy dye injection studies. His images indicate that the dye is in the subarachnoid rather than the subdural compartment and his autopsy description of less than 5 ml of blood in the subdural space where “nearly all the parasagittal bridging veins were completely torn” suggests that his methods are unreliable. Unless the BV are visualised before brain removal, artefactual rupture at autopsy cannot be excluded.

Not only are there no convincing pathological examples of BV rupture associated with thin film subdural bleeding but also there are physiological and anatomical objections to this hypothesis. Bridging veins are few in number—about 8–11 each side—and carry high blood flow (Fig. 4). In a 6-month baby, nearly 260 ml of blood flows into the dural sinuses per minute and the majority of cortical blood flows via the parasagittal BV into the superior sagittal sinus, where flow rate is 9.2 cm per second in the infant [90, 160, 178]. It is clear that rupture of even a single BV will cause massive space occupying clot, not a thin film, and the bleeding will be at least partly subarachnoid [37]. The suggestion that BV are weak at their dural junction was derived from studies of four elderly patients [181]; in fact, this junction consists of a smooth muscle sphincter which controls cerebral venous outflow when intracranial pressure is increased, maintaining the patency of the cortical veins [7, 141, 166, 183].

There appear to be circumstances when large cortical veins may ooze blood. Cushing observed that subdural haemorrhage “may occur when too great strain has been put upon the vessels by the profound venous stasis of postpartum asphyxiation; just as in later months they may rupture under the passive congestion brought about by a paroxysm of whooping-cough or a severe convulsion” [33]. Imaging and pathological observation support the suggestion of venous leakage under tension; radiological observations occasionally show small bleeds associated with BV crossing dilated extracerebral spaces and pathology of severely congested and thrombosed surface veins shows leakage of red cells across them into the subarachnoid space. In both circumstances, the bleeding is predominantly subarachnoid. (Fig. 5)

The dura

Intradural bleeding is common in the young infant and an almost universal finding at foetal and neonatal autopsy [28, 29, 59, 144, 149]. The anatomical and physiological basis for intradural bleeding in the infant has been discussed in detail [98]. In understanding the propensity for the infant dura to bleed, it must be borne in mind that the dura is not just a tough fibrous membrane providing physical support for the brain but it is the route of all venous outflow from the brain, via the dural sinuses.

The dura has two communicating vascular networks: the meningeal arteries, veins, and outermost periosteal plexus, which are superficially located, and the vascular plexus located between its periosteal and the meningeal leaflets, the remnant of a much more extensive network of the foetal dura [14]. The dense innervation of the dura is most abundant on the intradural sinuses and blood vessels. The dura also contains rounded fluid channels which may have a role in CSF uptake or monitoring [50, 120] and appears to be related both to age and to frequency and extent of intradural bleeding [148].

The dura at birth is very different from the dura after the end of the first year of life. At birth, the structure is of loosely arranged cells with a little collagen and the dural vascular plexuses and innervation are very much more extensive than in later life (unpublished observations); arachnoid granulations are not formed until about 7 months of postnatal life [14, 118]. These continuing developmental features may all influence the predisposition of the young baby to dural haemorrhage in the first months of life.

The posterior falx and tentorium are frequent sites of bleeding in both the foetus and the young infant dying of natural causes [28, 29]. These are also the sites of the posterior interhemispheric haemorrhage, originally regarded as characteristic of SBS. This radiological sign is most likely to be due to intradural bleeding or congestion of the abundant venous sinuses which are part of the normal anatomy in this age group. It is impossible for either MRI or CT scans to distinguish between intradural and thin film subdural bleeding. Figure 6 illustrates the case of a baby who collapsed with brain swelling and febrile convulsions. Thin film interhemispheric haemorrhage was identified as subdural bleeding on scan but at autopsy the blood was entirely intrafalcine. The detailed anatomy of the infant dura questions the validity of the belief that posterior interhemispheric bleeding is evidence of bridging vein rupture.

When it is extensive, intradural blood can almost always be seen seeping onto the subdural surface. Flecks of Perl’s positive material, often extracellular and close to the walls of the sinuses may represent older, or birth-related bleeding (Fig. 3).

Subdural bleeding is often seen adjacent to the lateral recesses of the superior sagittal sinus at the vertex [170], the sites where arachnoid granulations will develop towards the end of the first year of life. The functions of arachnoid granulations remain unclear; their role in resorption of CSF under normal conditions had long been disputed [34]; there is ample evidence that CSF is resorbed from cranial and spinal nerve roots [80, 186]. Arachnoid granulations have mechanoreceptors and may monitor CSF pressure, or act as valves preventing reflux of blood from the sinus into the CSF compartment. As young infants have no arachnoid granulations, venous blood from the superior sagittal sinus or the lateral lacunae may reflux into the dura and seep into the subdural compartment [148, 175].

An important and almost invariably overlooked part of the clinical history in babies presenting with the triad is a prolonged period of hypoxia, often 30 min or more between the baby being found collapsed and arriving in hospital and receiving advanced resuscitation. This sequence sets babies with the triad apart from cot death babies who are, by definition, found dead and have no pathology or intracranial bleeding. Prolonged hypoxia and resuscitation have been shown to be significantly associated with retinal haemorrhages [102] and may also explain the encephalopathy in babies with the triad. Experimental models of reperfusion injury confirm that longer periods of ischaemia cause greater small vessel damage and breakdown of the blood-brain barrier, exacerbated by resuscitation and reperfusion [97, 137].

Geddes proposed that in some infants with fatal head injury, the combination of severe hypoxia, brain swelling and raised central venous pressure is the cause of dural and retinal haemorrhage [59]. Geddes was not the first to make this observation; it had already been made by Cushing in 1905. There has been a tendency, notably in the Courts, to oversimplify this hypothesis to assume that hypoxia alone is a cause of subdural haemorrhage. This is misleading, as the physiological consequences of hypoxia are more complex. Subsequent research demonstrates an association between hypoxia and dural bleeding in young infants [28, 29]. Byard [16] found no SDH in hypoxic infants, but retrospective review of autopsy reports is unreliable in detecting small volume bleeds [22]. Hurley [74], in a retrospective autopsy and imaging study, found only one subdural haemorrhage and two intradural bleeds in 47 babies up to 4 years of age. Data regarding duration of hypoxia and resuscitation were incomplete. Neurological outcome is known to be far worse in children suffering out-of-hospital compared with in-hospital cardiac arrests, probably due to prolonged hypoxia and less effective CPR in the former group. [100].

A degree of subdural bleeding is extremely common after birth and seen on imaging in up to 46% of asymptomatic neonates after normal, instrumental and caesarean delivery [96, 132, 177]. More cases of SDH would be expected among symptomatic infants [24]. Two MRI studies have followed a total of just 27 babies with birth-related bleeding with repeated scans at 1–3 months of age. One baby developed further subdural bleeding [132, 177]. Due to the very small numbers used in these studies compared with the overall frequency of birth-related bleeding, meaningful interpretation is difficult and we have no good data on the natural history of birth-related SDH. It is obvious that most heal without any significant morbidity, although birth-related bleeding has been shown to be the cause of between 14 and 17% of infant chronic subdural haemorrhage [4, 69].

Studies of later onset infant subdural haemorrhage show that untreated small volume bleeds develop into chronic fluid collections. Loh [95] found chronic collections between 15 and 80 days after onset, the mean being 28 days. Hwang [75] described three cases of accidental subdural haemorrhage which resolved in CT scans only to reappear up to 111 days later. It seems likely that birth-related subdural haemorrhage will behave similarly.

Dating healing subdural haemorrhage by pathology alone is difficult and cannot constitute reliable evidence of the timing of an injury. In clinical practice, it is important to take into account the entire clinical history and the other clinical and pathological findings. Several guidelines for timing the cellular reactions to subdural haemorrhage are published [92, 108, 114].

The neuropathologist needs to be aware that there may not always be a clinical history to indicate pre-existing subdural haemorrhage. Chronic SDH can be extremely difficult to diagnose in infants and, unless specifically sought, the diagnosis is readily missed. Symptoms are non-specific and are sometimes purely systemic mimicking gastroenteritis, malnutrition or bronchopneumonia [68]; “Most often, the infant’s history includes failure to gain weight; refusal of feedings followed by frequent episodes of vomiting, some of which might be projectile; irritability; progressive enlargement of the head; and, ultimately, a seizure” [108].

Obstetric and birth records and brain scans should be reviewed. Head circumference charts are critical for the identification of extra-axial fluid and blood collections in life [184] and should be consulted in considering the possibility of pre-existing subdural haemorrhage.

The terminology of these entities is confused with no clear distinction between chronic subdural haemorrhagic collections, subdural hygromas and effusions. Subdural effusions may be xanthochromic or haemorrhagic and may evolve into frank subdural haematomas [19, 122]. Conversely, acute subdural haemorrhages may evolve into clear or xanthochromic protein-rich fluid collections or hygromas. The primary mechanism for the formation of hygromas remains unknown, it has been suggested that any pathologic condition at the dural border layer; fresh bleeding, lysis of pre-existing bleeding, inflammation or exudation from dural vessels can lead to effusion and fluid accumulation [48, 176].

Large fluid collections around the infant brain may be identified in otherwise normal babies and are usually self-limiting. Males outnumber females by 2–1. The causes are not known and include abnormalities in growth rates of the brain, the skull or the surrounding membranes, immaturity in the mechanisms of cerebrospinal fluid production and resorption, and old subdural bleeding. The many names, wide range of associated clinical findings and many aetiological hypotheses underscore the heterogeneity of the condition [60, 185]. Extra-axial fluid collections, whatever their cause, may predispose to SDH. Fresh bleeding into large extracerebral fluid collections after no, or only minor, trauma has been described [76, 107, 169]. Pittman [123] warned that acute SDH in the context of such a fluid collection around the brain could not be taken as evidence of abuse. A proposed mechanism for bleeding into extracerebral collections is leakage from over-stretched bridging veins which cross them [121]. Evidence from imaging and microscopy suggests that large surface veins may leak when their walls are stretched. Figure 5 demonstrates red cells passing between the cells of a congested and thrombosed cortical vein wall into the subarachnoid space.

In adults, the most common causes of SAH are trauma and ruptured aneurysm [43]. Hypoxia is a particularly frequent cause in the young infant, as are trauma and venous and sinus thrombosis.

Subpial bleeding receives a little attention in the pathological literature and is not generally distinguished from subarachnoid haemorrhage. Larroche [89] considered the pathogenesis and clinical implications to be the same. Subpial bleeding can be mistaken both radiologically and pathologically for contusion and erroneously support a diagnosis of trauma.

Friede described subpial bleeding as representing 15% of perinatal intracranial haemorrhage [52]. He considered that the bleeding dissected through the superficial astrocytic foot processes, and was a variant of, subarachnoid bleeding due to respiratory distress syndrome. Lindenberg [94] described similar bleeding into the outer part of cortical layer 1 in babies under 5 months and Voigt [171] described it in adults. Superficial bleeding, assumed to be subpial or subarachnoid, is described in temporal pole haemorrhage (see below) [73, 89].

Subpial bleeding is macroscopically well circumscribed, and often seen at the edge of a gyrus. I have seen it in association with cortical vein thrombosis, beneath space occupying subdural haematoma and beneath fractures occurring during forceps delivery. Larroche [89] considered occlusion or compression of superficial veins as a potential mechanism. The cortical veins, unlike cortical arteries, have little or no leptomeningeal investment around them [188] and bleeding around their deep cortical tributaries can track directly into the subpial space. (see Figs. 11, 13).

The skull bones develop within the outer mesenchymal layer which forms both the periosteum and the outer leaflet of the dura [98]. As the cranial dura is so intimately associated with the periosteum epidural bleeding is uncommon except where there has been surgery or fracture. Old trauma may not be obvious and there may be no history; skull fractures are associated with normal delivery and low falls and may be asymptomatic in the neonate [41, 135]. Rarely, cranial epidural bleeding is seen beneath an intact skull bone and is considered to result from inbending of the pliable infant skull (ping-pong fracture) which tears off the periosteum in the absence of fracture.

The relationship of the spinal dura to the vertebral bones is quite different. There is a wide epidural space which contains fat and the epidural plexus. This extensive, valveless plexus communicates above with the cranial venous outflow and is responsible for cerebral venous return to the heart in the upright position [163]. It becomes massively congested when intracranial pressure is reduced or when intra-abdominal pressure is increased [156, 157, 179].

Spinal epidural bleeding has been described in infants who are thought to have suffered non-accidental injury and has been considered to be evidence of shaking [64, 117]. However, spinal epidural haemorrhage is common in infants dying from all causes and is not diagnostic of trauma [136], but is probably a response to physiological and pathological variations in intracranial pressure (Fig. 2d).

Eisenbrey [45] was the first to suggest that retinal haemorrhages in a child under 4 years of age suggest abuse. Caffey [17] was prescient when he wrote “The retinal lesions caused by shaking will undoubtedly become valuable signs in the diagnosis of subclinical inapparent chronic subdural hematoma, and also become a productive screening test for the prevalence of whiplash dependent mental retardation and other types of so-called idiopathic brain damage”. Unilateral or bilateral retinal and vitreous haemorrhages, retinal folds and retinoschisis are indeed regarded as characteristic of Shaken Baby syndrome, and are estimated to be present in 65–90% of cases [40, 93, 134].

Vinchon [168] highlighted a pitfall in the use of RH in the diagnosis of abuse: “The importance of an RH for the diagnosis of child abuse is well established; however, the evaluation of its incidence in child abuse is almost impossible because the diagnosis of child abuse is in great part based on the presence of an RH, providing a circularity bias”. A further stumbling block in ascertaining the real significance of RH in abusive as compared to other forms of injury is that they may only be sought where abuse is suspected. In a study of SDH, ophthalmological opinion was sought in 94 of 106 cases of suspected NAHI but in only a quarter of babies with SDH of any other cause [69]. In a meta-analysis of 1,283 children fundoscopy was performed in 670 cases of suspected NAHI and only 328 cases with all other causes of brain injury [99].

All aspects of intraocular haemorrhage have been shown to occur without shaking [11, 127, 174]. A detailed, but yet unpublished, autopsy study found natural diseases to greatly outnumber inflicted injury in association with RH in infants under 1 year of age [87]. Thus, there is no diagnostic ocular neuropathology for SBS.

The pathologist must bear in mind that the ocular pathology resulting from the initial insult may be considerably modified by prolonged hypoxia, resuscitation, reperfusion, and a variable period of life support and must be interpreted in the context of the clinical findings closest to the time of injury. Clinical recognition of RH depends on the examination by an experienced physician using pupillary dilatation [168]; the timing of examination is crucial, as RH extend after initial injury [61].

A central issue is the mechanism of RH. The experimentally verified hydraulic theory is that retinal bleeding results from alterations in intracranial, intrathoracic and intra-abdominal pressure and blood pressure [146]. Muller and Deck [113] concluded that intraocular and optic nerve sheath haemorrhages result from the transmission of intracranial pressure into the optic nerve sheath and retinal venous hypertension. The alternative theory is that shaking causes vitreo-retinal traction, which tears the retina from its connections, disrupting the integrity of the blood vessels of the eye [93]. Ommaya [119] considered it biomechanically improbable that the levels of force generated by shaking would damage the eye directly and that a sudden rise of intracranial pressure is more likely to cause bleeding than the “shaken eye” hypothesis.

Observations of unilateral RH with ipsilateral intracranial haemorrhage or brain swelling indicate that RH may be due to the transmission of raised pressure along the optic nerve, potentially obstructing the central retinal vein [26, 61]. Pathology has not substantiated the theory of vitreo-retinal traction during shaking, but implicates a secondary phenomenon due to raised intracranial pressure or venous stasis and leakage from retinal vessels [46] This hypothesis is consistent with findings in the brain, whose capillary structure and physiology resembles that of the retina, where parenchymal bleeding tends to be associated with venous obstruction and tissue compression (see below and Fig. 15).

A statistically significant relationship between retinal and optic nerve sheath haemorrhage and reperfusion, cardiopulmonary resuscitation (CPR) and cerebral oedema has been demonstrated [102].

The list of conditions which may cause an infant to develop the triad is exhaustive. Below are brief notes on the most common causes of the triad which I have encountered in my own clinical diagnostic and forensic practices. Many others are discussed elsewhere [8, 51].